chore(tui): add diff close help changeset

.changeset/calm-moves-cache.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Preserve prompt cache prefixes when sessions move between locations with unchanged instructions.

.changeset/device-authorization-url.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Fix OpenCode Console device authorization URLs when the server returns an origin-rooted verification path.

.changeset/diff-close-help.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/tui": patch
+---
+
+Show the configured close shortcut in the diff viewer help dialog.

.changeset/directory-page-heading.md

@@ -1,5 +0,0 @@
----
-"@opencode/core": patch
----
-
-Correct directory page headings when the read offset is zero.

.changeset/git-ignore-check.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Propagate fatal Git ignore-check errors during index refresh so best-effort snapshot capture returns unavailable instead of treating the check as having no matches.

.changeset/idempotent-prompt-reuse.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Prompt and synthetic inbox ID reuse is now idempotent: reusing an ID within the same Session succeeds and returns the first admission, ignoring the retried payload, metadata, and delivery mode. Previously reuse with a differing payload failed with a conflict. Cross-Session and cross-type reuse still fail, and control items keep their operation-specific conflict behavior.

.changeset/model-content-names.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Rename the write, patch, and question tool formatting helpers from `toModelOutput` to `toModelContent` to match the result field they populate. Direct imports of these helpers must use the new name; generated content and declared machine output are unchanged.

.changeset/model-package-name.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": minor
+---
+
+Rename ModelResolver.supported to ModelResolver.hasPackage. Consumers of the old export must update the name; the predicate remains Boolean(model.package), checking only whether a catalog model declares a provider package, not whether it can be loaded. Default-model selection behavior is unchanged.

.changeset/nested-instruction-reinjection.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Nested AGENTS.md instructions are re-injected after compaction. Previously the in-memory dedup claim outlived the synthetic message that compaction dropped from model-visible history, so nested instructions were silently lost for the rest of the process lifetime. The claim now only guards in-flight loads; the synthetic message metadata in durable history is the sole lasting ledger, so any history truncation (compaction, revert) self-heals on the next read in that subtree.

.changeset/promise-plugin-adapter.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/plugin": patch
+---
+
+Derive Promise plugin API request and response conversion from the canonical protocol schemas.

.changeset/quiet-forms-replace-questions.md

@@ -0,0 +1,8 @@
+---
+"@opencode-ai/core": minor
+"@opencode-ai/schema": minor
+"@opencode-ai/protocol": minor
+"@opencode-ai/client": minor
+---
+
+Remove the unused question request API and use session forms for question tool interactions.

.changeset/rename-undo.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Include both paths of renamed files in new snapshot change lists so undo restores the original file instead of only deleting the renamed file.

.changeset/share-session-model-request.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Apply shared Session model-request preparation to transient generation.

.changeset/shell-permission-scan.md

@@ -0,0 +1,9 @@
+---
+"@opencode-ai/core": minor
+"@opencode-ai/schema": minor
+"@opencode-ai/protocol": minor
+"@opencode-ai/client": minor
+---
+
+Add an opt-in portable shell permission scanner. Opaque commands use normal shell authorization without inferring
+external directories, while the default tree-sitter path remains unchanged.

.changeset/shell-scanner-compatibility.md

@@ -0,0 +1,10 @@
+---
+"@opencode-ai/core": patch
+---
+
+Make the experimental portable shell scanner authoritative, with no Tree-sitter
+fallback. Scan common Bash and PowerShell control flow, heredocs, functions,
+expressions, quoting, and substitutions natively. Preserve existing redirect and
+declaration permission matching, and make PowerShell saved approvals cover the
+original command spelling. Parser failures remain visible without changing the
+permission engine. The default Tree-sitter path is unchanged.

.changeset/skill-directory-ids.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Use parent directory names for SKILL.md IDs so individually downloaded skills do not overwrite each other.

.changeset/steer-resume-simplified.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Simplify interrupt continuation: the steer-scoped resume decision now lives in SessionExecution as a post-cleanup inbox check, and the run coordinator drops its continuation state machine. Wakes arriving during cancellation cleanup now restart a normal full drain, and interrupting an idle session with continue now resumes pending steering input. Recovery-applied moves now end with the same full wake as inbox-admitted moves, retrying any stranded inbox work at the new location. Interrupting with continue now also resumes a next-in-line control item: between-turn manual compaction and moves run under any drain scope, while queued prompts remain parked.

.changeset/terminal-form-recovery.md

@@ -0,0 +1,6 @@
+---
+"@opencode-ai/client": patch
+"@opencode-ai/plugin": patch
+---
+
+Add form reply and cancellation operations that reconcile terminal forms in the local TUI projection.

.changeset/unify-request-envelope.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Title generation and compaction summaries now build their model requests through the shared session request boundary, gaining unsupported-media filtering and image bounds while explicitly opting out of session context hooks: plugins that shape the agent conversation do not observe title or compaction requests. Title requests gain the fork-aware session prompt cache key, and compaction summaries in forked sessions reuse the fork root's prompt cache key instead of the fork's own.

.changeset/websocket-request-hooks.md

@@ -0,0 +1,6 @@
+---
+"@opencode-ai/plugin": patch
+"@opencode-ai/core": patch
+---
+
+Add transport-neutral Session model request hooks and provider-scoped hook registration so eligible OpenAI Responses requests can prefer WebSocket without bypassing HTTP-only middleware.

.gitattributes

@@ -2,4 +2,3 @@ packages/core/migration/**/snapshot.json linguist-generated
 packages/core/src/database/migration.gen.ts linguist-generated
 packages/core/src/models-dev/snapshot.txt linguist-generated
 packages/core/src/**/*.txt text eol=lf
-packages/httpapi-codegen/test/generated/*.ts text eol=lf

.github/workflows/deploy.yml

@@ -5,7 +5,6 @@ on:
     branches:
       - dev
       - production
-      - beta
   workflow_dispatch:
 
 concurrency: ${{ github.workflow }}-${{ github.ref }}
@@ -16,7 +15,7 @@ permissions:
 
 jobs:
   deploy:
-    if: github.repository == 'anomalyco/opencode' && (github.ref_name == 'dev' || github.ref_name == 'production' || github.ref_name == 'beta')
+    if: github.repository == 'anomalyco/opencode' && (github.ref_name == 'dev' || github.ref_name == 'production')
     runs-on: ubuntu-latest
     environment: ${{ github.ref_name }}
     steps:
@@ -29,7 +28,6 @@ jobs:
           node-version: "24"
 
       - uses: aws-actions/configure-aws-credentials@7474bc4690e29a8392af63c5b98e7449536d5c3a # v4.3.1
-        if: github.ref_name != 'beta'
         with:
           role-to-assume: ${{ vars.AWS_DEPLOY_ROLE_ARN }}
           role-session-name: opencode-${{ github.run_id }}

.github/workflows/publish.yml

@@ -47,7 +47,7 @@ jobs:
       - uses: ./.github/actions/setup-bun
 
       - name: Deploy update service
-        if: github.ref_name == 'v2'
+        if: github.ref_name == 'v2' || github.ref_name == 'beta'
         working-directory: packages/updates
         run: bun run deploy
         env:
@@ -91,7 +91,7 @@ jobs:
 
       - uses: ./.github/actions/setup-bun
         with:
-          bun-version: 1.4.2
+          bun-version: 1.4.0
 
       - name: Setup git committer
         id: committer
@@ -113,7 +113,7 @@ jobs:
         id: build
         run: ./packages/cli/script/build.ts ${{ (github.ref_name == 'beta' && '--sourcemaps') || '' }}
         env:
-          BUN_COMPILE_RELEASE: bun-v1.4.2
+          BUN_COMPILE_RELEASE: bun-v1.4.0
           OPENCODE_VERSION: ${{ needs.version.outputs.version }}
           OPENCODE_RELEASE: ${{ needs.version.outputs.release }}
           GH_REPO: ${{ needs.version.outputs.repo }}
@@ -417,6 +417,7 @@ jobs:
       - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
 
       - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+        if: github.ref_name == 'beta'
         with:
           name: opencode-preview-cli
           path: packages/cli/dist
@@ -479,7 +480,7 @@ jobs:
           OPENCODE_VERSION: ${{ needs.version.outputs.version }}
           OPENCODE_CHANNEL: ${{ (github.ref_name == 'beta' && 'beta') || 'prod' }}
           OPENCODE_CLI_TARGET: ${{ matrix.settings.target }}
-          OPENCODE_CLI_DIST: ${{ github.workspace }}/packages/cli/dist
+          OPENCODE_CLI_DIST: ${{ (github.ref_name == 'beta' && format('{0}/packages/cli/dist', github.workspace)) || '' }}
 
       - name: Build
         run: bun run build

.github/workflows/test.yml

@@ -49,7 +49,7 @@ jobs:
             echo "app=true" >> "$GITHUB_OUTPUT"
             exit 0
           fi
-          bun x turbo@2.10.2 ls --affected --filter=@opencode/app --output=json > affected.json
+          bun x turbo@2.10.2 ls --affected --filter=@opencode-ai/app --output=json > affected.json
           bun -e 'const result = await Bun.file("affected.json").json(); console.log(`app=${result.packages.count > 0}`)' >> "$GITHUB_OUTPUT"
 
   unit:
@@ -132,10 +132,10 @@ jobs:
         timeout-minutes: 15
         run: |
           if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
-            bun turbo verify:package --filter=@opencode/sdk
+            bun turbo verify:package --filter=@opencode-ai/sdk
             exit 0
           fi
-          bun turbo verify:package --affected --filter=@opencode/sdk
+          bun turbo verify:package --affected --filter=@opencode-ai/sdk
         env:
           TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
           TURBO_SCM_HEAD: ${{ github.sha }}

.opencode/tool/github-pr-search.ts

@@ -1,5 +1,5 @@
 /// <reference path="../env.d.ts" />
-import { tool } from "@opencode/plugin"
+import { tool } from "@opencode-ai/plugin"
 async function githubFetch(endpoint: string, options: RequestInit = {}) {
   const response = await fetch(`https://api.github.com${endpoint}`, {
     ...options,

.opencode/tool/github-triage.ts

@@ -1,5 +1,5 @@
 /// <reference path="../env.d.ts" />
-import { tool } from "@opencode/plugin"
+import { tool } from "@opencode-ai/plugin"
 
 const TEAM = {
   tui: ["kommander", "simonklee"],

.oxlintrc.json

@@ -41,5 +41,11 @@
     // Warn when spreading non-plain objects (Headers, class instances, etc.)
     "typescript/no-misused-spread": "warn"
   },
+  "options": {
+    "typeAware": true
+  },
+  "options": {
+    "typeAware": true
+  },
   "ignorePatterns": ["**/node_modules", "**/dist", "**/.build", "**/.sst", "**/*.d.ts", "**/sdk.gen.ts"]
 }

AGENTS.md

@@ -1,9 +1,8 @@
 - After changing the public Protocol or Server `HttpApi`, run `bun run generate` from `packages/client`. Do not edit generated client files directly.
 - Keep runtime dependencies directed from Schema to Core and Protocol, then from Core and Protocol to Server. Client runtime code may depend on Schema and Protocol but never Core or Server; `sdk` composes Client, Core, and Server.
 - Current implementation changes belong in `packages/core`, `packages/cli`, `packages/server`, `packages/protocol`, `packages/schema`, and related generated client surfaces when required.
-- This repository does not use Changesets. Do not add `.changeset` files; follow the existing release workflow instead.
 - The default branch in this repo is `v2`.
-- Default new branches and worktrees to `v2`, or `origin/v2` when the local `v2` ref is unavailable, and default pull requests to target `v2`. Use another base or target branch when the requester explicitly instructs it.
+- Base all new branches and worktrees on `v2`, or `origin/v2` when the local `v2` ref is unavailable. Do not base them on `dev`.
 - Local `main` ref may not exist; use `v2` or `origin/v2` for diffs.
 
 ## Live V2 TUI Testing
@@ -46,7 +45,6 @@ Examples: `fix(tui): simplify thinking toggle styling`, `docs: update contributi
 ### General Principles
 
 - Keep things in one function unless composable or reusable
-- Validate unknown values once at the boundary that owns them. Pass typed values inward instead of repeating `typeof value === "object"` and property-existence checks. Do not defensively revalidate values already guaranteed by a schema, constructor, or internal type.
 - Do not extract single-use helpers preemptively. Inline the logic at the call site unless the helper is reused, hides a genuinely complex boundary, or has a clear independent name that improves the caller.
 - Before adding complexity for a speculative or vanishingly unlikely race or security edge case, explain the concrete failure mode, likelihood, and complexity cost to the user and get their buy-in. Do not silently expand scope for theoretical robustness.
 - Avoid `try`/`catch` where possible
@@ -84,9 +82,9 @@ const { a, b } = obj
 ### Imports
 
 - Never alias imports. Do not use `import { foo as bar } from "..."` or renamed imports like `resolve as pathResolve`.
-- Never use type-position `import("...")` references such as `Schema.declare<import("@opencode/plugin/effect/plugin").Plugin["effect"]>`. Only when two imports genuinely collide on a name and no other option exists, an aliased type import (`import type { Plugin as PluginDefinition } from "..."`) is permitted as a last resort — still strongly preferred not to.
+- Never use type-position `import("...")` references such as `Schema.declare<import("@opencode-ai/plugin/effect/plugin").Plugin["effect"]>`. Only when two imports genuinely collide on a name and no other option exists, an aliased type import (`import type { Plugin as PluginDefinition } from "..."`) is permitted as a last resort — still strongly preferred not to.
 - Never use star imports. Do not use `import * as Foo from "..."` or `import type * as Foo from "..."`.
-- If a namespace-style value is needed, import the module's own exported namespace by name, for example `import { Project } from "@opencode/core/project"`, then reference `Project.ID`.
+- If a namespace-style value is needed, import the module's own exported namespace by name, for example `import { Project } from "@opencode-ai/core/project"`, then reference `Project.ID`.
 - Prefer dynamic imports for heavy modules that are only needed in selected code paths, especially in startup-sensitive entrypoints. Destructure dynamic import bindings near the top of the narrowest scope that needs them so they read like normal imports. Avoid inline chains such as `await import("./module").then((mod) => mod.value())` or `(await import("./module")).value()`. Keep branch-specific imports inside the branch that needs them to preserve lazy loading.
 
 ### Variables
@@ -183,7 +181,7 @@ const table = sqliteTable("session", {
 - Keep `SessionRunner`, model resolution, tool registry, permissions, and filesystem Location-scoped. Omitted `Location.workspaceID` means implicit-local placement; explicit workspace identity remains reserved for future placement semantics.
 - Preserve one explicit `llm.stream(request)` call per Physical Attempt and reload projected history before durable continuation. A logical Step may use generic pre-output retries, one full-context retry after continuation rejection, incomplete-stream continuation, or one overflow-compaction rebuild. Generic retries retain the logical step number and do not consume another agent-step allowance. Do not delegate orchestration to an in-memory tool loop.
 - Keep local Session drains process-local until clustering is implemented. `SessionRunCoordinator` joins explicit same-Session resumes, coalesces prompt wakeups, and allows different Sessions to run concurrently. A write-ahead execution claim marks a process-local busy period for restart recovery: terminal completion, failure, or user interruption releases it, while shutdown interruption and process death preserve it. Startup recovery resumes claimed top-level Sessions with durable per-execution attempt accounting. The claim is a recovery marker, not clustered ownership, fencing, or an exactly-once guarantee.
-- Keep delivery vocabulary explicit. Prompts steer by default. At safe step boundaries, steered compaction takes priority up to the first steered move control; other steers retain enqueue order. At an idle boundary, steers take priority; otherwise exactly one queued item delivers before the runner reevaluates continuation. Inbox items may be cancelled or changed between queue and steer before delivery. Promoting new user input resets the selected agent's step allowance; a batch of steers resets it once.
+- Keep delivery vocabulary explicit. Prompts steer by default. Steers deliver in enqueue order at safe step boundaries, stopping before compaction or move control items. At an idle boundary, steers take priority; otherwise exactly one queued item delivers before the runner reevaluates continuation. Inbox items may be cancelled or changed between queue and steer before delivery. Promoting new user input resets the selected agent's step allowance; a batch of steers resets it once.
 - One step is one logical LLM call; its durable record covers only the model-visible span. Do not write "provider turn", and do not use bare "turn" for a single call: "turn" is reserved for the future assistant-turn unit containing all steps from prompt promotion until the session would go idle.
 - Keep event replay ownership separate from clustered Session execution ownership.
 - Keep the Instructions algebra and built-ins in `src/instructions`; keep instruction producers with their observed domains, and keep Session History selection plus `InstructionState` and `InstructionEntry` persistence Session-owned. `InstructionDiscovery` observes ambient global and upward-project instructions. The runner composes built-ins, discovery, guidance, and entries explicitly in `loadInstructions`; there is no instruction registry.

bunfig.toml

@@ -2,7 +2,7 @@
 exact = true
 # Only install newly resolved package versions published at least 3 days ago.
 minimumReleaseAge = 259200
-minimumReleaseAgeExcludes = ["@ai-sdk/amazon-bedrock", "@ai-sdk/anthropic", "@brendonovich/vite-plugin-opencode", "@opencode/sdk", "@opencode-ai/pty", "@opencode-ai/pty-darwin-arm64", "@opencode-ai/pty-darwin-x64", "@opencode-ai/pty-linux-arm64-gnu", "@opencode-ai/pty-linux-arm64-musl", "@opencode-ai/pty-linux-x64-gnu", "@opencode-ai/pty-linux-x64-musl", "@opentui/core", "@opentui/core-darwin-arm64", "@opentui/core-darwin-x64", "@opentui/core-linux-arm64", "@opentui/core-linux-arm64-musl", "@opentui/core-linux-x64", "@opentui/core-linux-x64-musl", "@opentui/core-win32-arm64", "@opentui/core-win32-x64", "@opentui/keymap", "@opentui/solid", "opentui-spinner", "gitlab-ai-provider", "opencode-gitlab-auth", "@ff-labs/fff-node", "@ff-labs/fff-bun", "@ff-labs/fff-bin-darwin-arm64", "@ff-labs/fff-bin-darwin-x64", "@ff-labs/fff-bin-linux-arm64-gnu", "@ff-labs/fff-bin-linux-arm64-musl", "@ff-labs/fff-bin-linux-x64-gnu", "@ff-labs/fff-bin-linux-x64-musl", "@ff-labs/fff-bin-win32-arm64", "@ff-labs/fff-bin-win32-x64", "@pierre/diffs", "@pierre/theming", "app-builder-lib", "dmg-builder", "electron", "electron-builder", "electron-publish", "blume", "mermaid"]
+minimumReleaseAgeExcludes = ["@ai-sdk/amazon-bedrock", "@ai-sdk/anthropic", "@brendonovich/vite-plugin-opencode", "@opencode-ai/sdk", "@opencode-ai/pty", "@opencode-ai/pty-darwin-arm64", "@opencode-ai/pty-darwin-x64", "@opencode-ai/pty-linux-arm64-gnu", "@opencode-ai/pty-linux-arm64-musl", "@opencode-ai/pty-linux-x64-gnu", "@opencode-ai/pty-linux-x64-musl", "@opentui/core", "@opentui/core-darwin-arm64", "@opentui/core-darwin-x64", "@opentui/core-linux-arm64", "@opentui/core-linux-arm64-musl", "@opentui/core-linux-x64", "@opentui/core-linux-x64-musl", "@opentui/core-win32-arm64", "@opentui/core-win32-x64", "@opentui/keymap", "@opentui/solid", "opentui-spinner", "gitlab-ai-provider", "opencode-gitlab-auth", "@ff-labs/fff-node", "@ff-labs/fff-bun", "@ff-labs/fff-bin-darwin-arm64", "@ff-labs/fff-bin-darwin-x64", "@ff-labs/fff-bin-linux-arm64-gnu", "@ff-labs/fff-bin-linux-arm64-musl", "@ff-labs/fff-bin-linux-x64-gnu", "@ff-labs/fff-bin-linux-x64-musl", "@ff-labs/fff-bin-win32-arm64", "@ff-labs/fff-bin-win32-x64", "@pierre/diffs", "@pierre/theming", "app-builder-lib", "dmg-builder", "electron-builder", "electron-publish", "blume"]
 
 [test]
 root = "./do-not-run-tests-from-root"

infra/app.ts

@@ -1,5 +1,4 @@
 import { domain } from "./stage"
-import { createWebApp } from "./webapp"
 
 const GITHUB_APP_ID = new sst.Secret("GITHUB_APP_ID")
 const GITHUB_APP_PRIVATE_KEY = new sst.Secret("GITHUB_APP_PRIVATE_KEY")
@@ -60,4 +59,11 @@ new sst.cloudflare.x.Astro("Web", {
   },
 })
 
-createWebApp("app." + domain)
+new sst.cloudflare.StaticSite("WebApp", {
+  domain: "app." + domain,
+  path: "packages/app",
+  build: {
+    command: "bun turbo build",
+    output: "./dist",
+  },
+})

infra/console.ts

@@ -235,7 +235,6 @@ const bucketNew = new sst.cloudflare.Bucket("ZenDataNew")
 const DISCORD_INCIDENT_WEBHOOK_URL = new sst.Secret("DISCORD_INCIDENT_WEBHOOK_URL")
 const AWS_SES_ACCESS_KEY_ID = new sst.Secret("AWS_SES_ACCESS_KEY_ID")
 const AWS_SES_SECRET_ACCESS_KEY = new sst.Secret("AWS_SES_SECRET_ACCESS_KEY")
-const ENTERPRISE_SALES_INBOX_EMAIL = new sst.Secret("ENTERPRISE_SALES_INBOX_EMAIL")
 
 const SALESFORCE_CLIENT_ID = new sst.Secret("SALESFORCE_CLIENT_ID")
 const SALESFORCE_CLIENT_SECRET = new sst.Secret("SALESFORCE_CLIENT_SECRET")
@@ -264,7 +263,6 @@ new sst.cloudflare.x.SolidStart("Console", {
     EMAILOCTOPUS_API_KEY,
     AWS_SES_ACCESS_KEY_ID,
     AWS_SES_SECRET_ACCESS_KEY,
-    ENTERPRISE_SALES_INBOX_EMAIL,
     SALESFORCE_CLIENT_ID,
     SALESFORCE_CLIENT_SECRET,
     SALESFORCE_INSTANCE_URL,

infra/webapp.ts

@@ -1,18 +0,0 @@
-export function createWebApp(domain: string) {
-  return new sst.cloudflare.StaticSite("WebApp", {
-    domain,
-    path: "packages/app",
-    environment:
-      $app.stage === "beta"
-        ? {
-            OPENCODE_CHANNEL: "beta",
-            VITE_SENTRY_ENVIRONMENT: "beta",
-          }
-        : undefined,
-    build: {
-      // Preserve Sentry credentials and run source-map uploads on every deployment.
-      command: "bun run build",
-      output: "./dist",
-    },
-  })
-}

install

@@ -165,30 +165,22 @@ else
         exit 1
     fi
 
-    package_scope="@opencode"
     if [ -z "$requested_version" ]; then
-        metadata=$(curl -fsSL https://opencode.ai/update/api/beta/cli/npm || true)
+        metadata=$(curl -fsSL https://registry.npmjs.org/@opencode-ai%2fcli/beta || true)
         specific_version=$(echo "$metadata" | sed -n 's/.*"version":"\([^"]*\)".*/\1/p')
-        package=$(echo "$metadata" | sed -n 's/.*"package":"\([^"]*\)".*/\1/p')
 
-        if [ -z "$specific_version" ] || [ -z "$package" ]; then
+        if [ -z "$specific_version" ]; then
             echo -e "${RED}Failed to fetch version information${NC}"
             exit 1
         fi
-        package_scope="${package%/cli}"
     else
         # Strip leading 'v' if present
         requested_version="${requested_version#v}"
         specific_version=$requested_version
     fi
 
-    package_name="$package_scope/cli-$target"
-    http_status=$(curl -s -o /dev/null -w "%{http_code}" "https://registry.npmjs.org/$package_scope%2fcli-$target/$specific_version" || true)
-    # Older clients install the minimum release before they can migrate package names.
-    if [ "$http_status" = "404" ] && [ -n "$requested_version" ]; then
-        package_name="@opencode-ai/cli-$target"
-        http_status=$(curl -s -o /dev/null -w "%{http_code}" "https://registry.npmjs.org/@opencode-ai%2fcli-$target/$specific_version" || true)
-    fi
+    package_name="@opencode-ai/cli-$target"
+    http_status=$(curl -s -o /dev/null -w "%{http_code}" "https://registry.npmjs.org/@opencode-ai%2fcli-$target/$specific_version" || true)
     if [ "$http_status" = "404" ]; then
         echo -e "${RED}Error: Version ${specific_version} is not available for $target${NC}"
         echo -e "${MUTED}Available versions: https://www.npmjs.com/package/$package_name?activeTab=versions${NC}"

nix/desktop.nix

@@ -4,7 +4,7 @@
   bun,
   nodejs,
   darwin,
-  callPackage,
+  electron_41,
   makeWrapper,
   writableTmpDirAsHomeHook,
   autoPatchelfHook,
@@ -13,7 +13,7 @@
   opencode,
 }:
 let
-  electron = callPackage ./electron.nix { };
+  electron = electron_41;
 in
 stdenv.mkDerivation (finalAttrs: {
   pname = "opencode-desktop";
@@ -48,7 +48,7 @@ stdenv.mkDerivation (finalAttrs: {
     desktopName = "OpenCode";
     exec = "opencode-desktop %U";
     icon = "ai.opencode.desktop";
-    # Electron derives X11 WM_CLASS from app.name.
+    # Electron 41 derives X11 WM_CLASS from app.name.
     startupWMClass = "OpenCode";
     categories = [ "Development" ];
   });
@@ -67,10 +67,12 @@ stdenv.mkDerivation (finalAttrs: {
     # https://github.com/electron/electron/issues/31121
     # mac builds use a .app bundle which doesnt have this issue
     + lib.optionalString stdenv.isLinux ''
-      substituteInPlace \
-        packages/desktop/src/main/windows/appearance.ts \
-        packages/desktop/src/main/service/desktop-cli.ts \
-        --replace-fail "process.resourcesPath" "'$out/opt/opencode-desktop/resources'"
+      BASE_PATH=packages/desktop
+      FILES=(src/main/windows.ts)
+      for file in "''${FILES[@]}"; do
+        substituteInPlace $BASE_PATH/$file \
+          --replace-fail "process.resourcesPath" "'$out/opt/opencode-desktop/resources'"
+      done
     '';
 
   preBuild = ''
@@ -87,11 +89,6 @@ stdenv.mkDerivation (finalAttrs: {
 
     cd packages/desktop
 
-    export OPENCODE_CLI_DIST="$TMPDIR/desktop-cli"
-    cli_package=$(bun -e 'import { getCurrentCli } from "./scripts/utils.ts"; console.log(getCurrentCli().package.replace("@opencode/", ""))')
-    mkdir -p "$OPENCODE_CLI_DIST/$cli_package/bin"
-    cp ${lib.getExe opencode} "$OPENCODE_CLI_DIST/$cli_package/bin/opencode2"
-
     bun run build
     npx electron-builder --dir \
       --config electron-builder.config.ts \

nix/electron.nix

@@ -1,13 +0,0 @@
-{ callPackage, path }:
-let
-  version = (builtins.fromJSON (builtins.readFile ../packages/desktop/package.json)).devDependencies.electron;
-in
-(callPackage (path + "/pkgs/development/tools/electron/binary/generic.nix") { }) version {
-  # Electron 42.10.1 SHASUMS256.txt; update with the desktop package version.
-  aarch64-linux = "20e68d6c4e47f3ebf59de7c6b1f8b8bec6a6ebda6a451132f9b465f3f13ce467";
-  x86_64-linux = "2452b27112d92387471fa2488aafac85d79ea3f2ee1216c0abd5150d6c12362b";
-  aarch64-darwin = "ac7194a3dfd81930ba35355c01620262c1254752859b42dcb8f4b9e4d174a871";
-  x86_64-darwin = "4489aba55477a0082266cb690db1c829503ba3338048599d8fd243953df37dab";
-  # fetchzip hashes the unpacked headers, not the release tarball.
-  headers = "sha256-4eUy3BZVvxTl7KUOsxio7769lL6ag/ecbeK+qLURWMI=";
-}

nix/hashes.json

@@ -1,8 +1,8 @@
 {
   "nodeModules": {
-    "x86_64-linux": "sha256-+E2chSxD9x139rXmeomoJMiNybSAAZstB2UiOBst8nE=",
-    "aarch64-linux": "sha256-Wsm3Q4+k1kzxP0aBv2MRJl2sW/q/7CD9GuWsd4J7xEo=",
-    "aarch64-darwin": "sha256-6lrSRpyGrN0FmQ3dXSaUERUQyW4IxjXCCeWi5oJ6GyM=",
-    "x86_64-darwin": "sha256-V7XSss7r0hkUsH92rtJhwuWEZXnemridoF6wCTbl8Lc="
+    "x86_64-linux": "sha256-agxF1Ab8P84tijHvw7jaG+9UZm/dJZhHw8CGNlQQqYc=",
+    "aarch64-linux": "sha256-Kg6CkhoEEsb+ZOGPN/rnNoirLdGSyQl6HU/Fy8t9Mqs=",
+    "aarch64-darwin": "sha256-SnxBpmcfPL2p3OYmSuG5Y/aFW8PeEOLhfQrxAs/53fs=",
+    "x86_64-darwin": "sha256-Dnxl1EcfiE7/rlHmPkko0BKc6hHVk6rQ/2h/PWuLfNo="
   }
 }

nix/node_modules.nix

@@ -31,7 +31,7 @@ stdenvNoCC.mkDerivation {
         ../package.json
         ../patches
         ../install # required by desktop build (cli.rs include_str!)
-        ../.github/TEAM_MEMBERS # required by @opencode/script
+        ../.github/TEAM_MEMBERS # required by @opencode-ai/script
       ]
     );
   };

package.json

@@ -5,10 +5,10 @@
   "version": "0.0.0",
   "private": true,
   "type": "module",
-  "packageManager": "bun@1.4.2",
+  "packageManager": "bun@1.3.14",
   "scripts": {
-    "dev": "bun run --cwd packages/cli src/index.ts",
-    "dev:live": "sh -c 'OPENCODE_TUI_CHANNEL=dev OPENCODE_PASSWORD=\"$(opencode2 service get password)\" exec bun run dev \"$@\" --server \"$(opencode2 service status)\"' --",
+    "dev": "bun run --cwd packages/cli --conditions=browser src/index.ts",
+    "dev:live": "OPENCODE_TUI_CHANNEL=dev OPENCODE_PASSWORD=\"$(opencode2 service get password)\" bun run dev --server \"$(opencode2 service status)\"",
     "dev:desktop": "bun --cwd packages/desktop dev",
     "dev:web": "bun --cwd packages/app dev",
     "dev:console": "ulimit -n 10240 2>/dev/null; bun run --cwd packages/console/app dev",
@@ -27,6 +27,7 @@
     "upgrade-opentui": "bun run script/upgrade-opentui.ts",
     "postinstall": "bun run --cwd packages/core fix-node-pty",
     "prepare": "husky",
+    "reserve-packages": "bun script/reserve-package-names.ts",
     "random": "echo 'Random script'",
     "sso": "aws sso login --sso-session=opencode --no-browser",
     "test": "echo 'do not run tests from root' && exit 1"
@@ -38,24 +39,24 @@
       "packages/stats/*"
     ],
     "catalog": {
-      "@effect/opentelemetry": "4.0.0-rc.112",
-      "@effect/platform-node": "4.0.0-rc.112",
-      "@effect/platform-node-shared": "4.0.0-rc.112",
-      "@effect/sql-sqlite-bun": "4.0.0-rc.112",
+      "@effect/opentelemetry": "4.0.0-rc.111",
+      "@effect/platform-node": "4.0.0-rc.111",
+      "@effect/platform-node-shared": "4.0.0-rc.111",
+      "@effect/sql-sqlite-bun": "4.0.0-rc.111",
       "@npmcli/arborist": "9.4.0",
-      "@types/bun": "1.4.0",
+      "@types/bun": "1.3.13",
       "@types/cross-spawn": "6.0.6",
       "@octokit/rest": "22.0.0",
       "@hono/standard-validator": "0.2.0",
       "@hono/zod-validator": "0.4.2",
-      "@opentui/core": "0.5.10",
-      "@opentui/keymap": "0.5.10",
-      "@opentui/solid": "0.5.10",
+      "@opentui/core": "0.5.8",
+      "@opentui/keymap": "0.5.8",
+      "@opentui/solid": "0.5.8",
       "@tanstack/solid-virtual": "3.13.37",
-      "@shikijs/stream": "4.4.3",
+      "@shikijs/stream": "4.2.0",
       "@standard-schema/spec": "1.1.0",
       "ulid": "3.0.1",
-      "@kobalte/core": "0.13.13",
+      "@kobalte/core": "0.13.11",
       "@corvu/drawer": "0.2.4",
       "@types/luxon": "3.7.1",
       "@types/node": "24.12.2",
@@ -66,16 +67,13 @@
       "@openauthjs/openauth": "0.0.0-20250322224806",
       "@pierre/diffs": "1.2.10",
       "opentui-spinner": "0.0.7",
-      "@solid-primitives/event-listener": "2.4.6",
-      "@solid-primitives/media": "2.3.6",
-      "@solid-primitives/resize-observer": "2.2.0",
-      "@solid-primitives/storage": "4.4.0",
+      "@solid-primitives/storage": "4.3.3",
       "@tailwindcss/vite": "4.1.11",
-      "diff": "8.0.4",
-      "dompurify": "3.4.14",
+      "diff": "8.0.2",
+      "dompurify": "3.3.1",
       "drizzle-kit": "1.0.0-rc.5-ab785fc",
       "drizzle-orm": "1.0.0-rc.5-169397b",
-      "effect": "4.0.0-rc.112",
+      "effect": "4.0.0-rc.111",
       "ai": "6.0.168",
       "cross-spawn": "7.0.6",
       "hono": "4.10.7",
@@ -83,27 +81,28 @@
       "fuzzysort": "3.1.0",
       "get-east-asian-width": "1.6.0",
       "luxon": "3.6.1",
-      "marked": "18.0.11",
+      "marked": "18.0.7",
       "marked-shiki": "1.2.1",
-      "remend": "1.3.1",
+      "remend": "1.3.0",
       "@playwright/test": "1.59.1",
       "semver": "7.7.4",
       "typescript": "5.8.2",
       "@typescript/native-preview": "7.0.0-dev.20251207.1",
       "zod": "4.1.8",
       "remeda": "2.26.0",
+      "resolve.exports": "2.0.3",
       "sst": "4.13.1",
-      "shiki": "4.4.3",
+      "shiki": "4.2.0",
       "solid-list": "0.3.0",
       "string-width": "7.2.0",
       "tailwindcss": "4.1.11",
-      "vite": "7.3.6",
+      "vite": "7.1.4",
       "@solidjs/meta": "0.29.4",
-      "@solidjs/router": "1.0.0",
+      "@solidjs/router": "0.15.4",
       "@solidjs/start": "https://pkg.pr.new/@solidjs/start@dfb2020",
-      "@sentry/solid": "10.71.0",
-      "@sentry/vite-plugin": "5.4.0",
-      "solid-js": "1.9.15",
+      "@sentry/solid": "10.36.0",
+      "@sentry/vite-plugin": "4.6.0",
+      "solid-js": "1.9.10",
       "solid-sonner": "0.3.1",
       "vite-plugin-solid": "2.11.10",
       "@lydell/node-pty": "1.2.0-beta.12"
@@ -128,8 +127,8 @@
   },
   "dependencies": {
     "@aws-sdk/client-s3": "3.933.0",
-    "@opencode/plugin": "workspace:*",
-    "@opencode/script": "workspace:*",
+    "@opencode-ai/plugin": "workspace:*",
+    "@opencode-ai/script": "workspace:*",
     "heap-snapshot-toolkit": "1.1.3",
     "typescript": "catalog:"
   },
@@ -146,6 +145,10 @@
     "esbuild",
     "node-pty",
     "protobufjs",
+    "tree-sitter",
+    "tree-sitter-bash",
+    "tree-sitter-powershell",
+    "web-tree-sitter",
     "electron"
   ],
   "overrides": {
@@ -155,9 +158,7 @@
     "@effect/platform-node-shared": "catalog:",
     "@types/bun": "catalog:",
     "@types/node": "catalog:",
-    "bun-types": "1.4.2",
-    "effect": "catalog:",
-    "solid-js": "catalog:"
+    "effect": "catalog:"
   },
   "patchedDependencies": {
     "@ai-sdk/openai-compatible@2.0.41": "patches/@ai-sdk%2Fopenai-compatible@2.0.41.patch",
@@ -165,7 +166,7 @@
     "@npmcli/agent@4.0.2": "patches/@npmcli%2Fagent@4.0.2.patch",
     "@silvia-odwyer/photon-node@0.3.4": "patches/@silvia-odwyer%2Fphoton-node@0.3.4.patch",
     "@standard-community/standard-openapi@0.2.9": "patches/@standard-community%2Fstandard-openapi@0.2.9.patch",
-    "solid-js@1.9.15": "patches/solid-js@1.9.15.patch",
+    "solid-js@1.9.10": "patches/solid-js@1.9.10.patch",
     "@ai-sdk/mistral@3.0.51": "patches/@ai-sdk%2Fmistral@3.0.51.patch",
     "gcp-metadata@8.1.2": "patches/gcp-metadata@8.1.2.patch",
     "pacote@21.5.0": "patches/pacote@21.5.0.patch",
@@ -173,8 +174,6 @@
     "@pierre/trees@1.0.0-beta.4": "patches/@pierre%2Ftrees@1.0.0-beta.4.patch",
     "@modelcontextprotocol/sdk@1.29.0": "patches/@modelcontextprotocol%2Fsdk@1.29.0.patch",
     "@tanstack/virtual-core@3.17.8": "patches/@tanstack%2Fvirtual-core@3.17.8.patch",
-    "@ff-labs/fff-bun@0.10.5": "patches/@ff-labs%2Ffff-bun@0.10.5.patch",
-    "ghostty-web@github:anomalyco/ghostty-web#83c0a07": "patches/ghostty-web@0.3.0.patch",
-    "vite@8.2.2": "patches/vite@8.2.2.patch"
+    "@ff-labs/fff-bun@0.10.5": "patches/@ff-labs%2Ffff-bun@0.10.5.patch"
   }
 }

packages/ai/AGENTS.md

@@ -81,7 +81,7 @@ export const route = Route.make({
 
 Route defaults are request-shaping defaults such as `headers`, `limits`, `generation`, `providerOptions`, and `http`. Endpoint host/query belongs on the route endpoint. Selected `LanguageModel` values carry only model id, provider id, and the configured route value. Model capability/catalog metadata lives outside this package; protocol support is enforced by request lowering and typed `AIError`s.
 
-The four-axis decomposition is the reason DeepSeek, TogetherAI, Cerebras, Baseten, Fireworks, and DeepInfra all reuse `OpenAIChat.protocol` verbatim — each provider owns a small `Route.make(...)` composition instead of a protocol clone. Bug fixes in one protocol propagate to every consumer of that protocol in a single commit.
+The four-axis decomposition is the reason DeepSeek, TogetherAI, Cerebras, Baseten, Fireworks, and DeepInfra all reuse `OpenAIChat.protocol` verbatim — each provider deployment is a 5-15 line `Route.make(...)` call instead of a 300-400 line route clone. Bug fixes in one protocol propagate to every consumer of that protocol in a single commit.
 
 When a provider supports multiple physical transports, selection remains execution policy below its semantic route. `OpenResponsesChannel.transport(...)` owns the provider-neutral Responses WebSocket concept: it prepares one final request, executes HTTP by default, strips WebSocket-disallowed fields, and passes a generic channel exchange to a per-call `WebSocketChannelExecutor` when supplied. Provider-specific Responses routes opt in with handshake and connection-age policy. `Route.streamPrepared` owns decoding and acknowledges channel completion only after successful full consumption.
 
@@ -115,16 +115,15 @@ Keep provider facades small and explicit:
 - Prefer `apiKey` as provider-specific sugar and `auth` as the explicit override; keep them mutually exclusive in provider option types with `ProviderAuthOption`.
 - Resolve `apiKey` → `Auth` with `AuthOptions.bearer(options, "<PROVIDER>_API_KEY")` (it honors an explicit `auth` override and falls back to `Auth.config(envVar)` so missing keys surface a typed `Authentication` error rather than a runtime crash).
 - Use separate top-level facades for products with different required setup, such as `CloudflareAIGateway` and `CloudflareWorkersAI`.
-- Give every named provider its own file and top-level export. Keep its endpoint, auth defaults, and route setup in that file. Compose shared protocols directly; do not nest named provider presets under generic compatible facades or keep their endpoints in a shared provider profile registry.
 
 `Provider.make(...)` remains available for simple static provider definitions, but new built-in providers should prefer plain configured facades unless a helper removes real duplication without adding runtime behavior.
 
 ### Provider Package Entrypoints
 
-Catalog-selected native providers use package-like export paths from `@opencode/ai`. They are internal entrypoints in one npm package, not separately published provider packages. Every entrypoint implements `ProviderPackage.Definition` and exposes `model(modelID, settings)`, where settings are serializable provider configuration plus common `headers`, `body`, and `limits` overlays.
+Catalog-selected native providers use package-like export paths from `@opencode-ai/ai`. They are internal entrypoints in one npm package, not separately published provider packages. Every entrypoint implements `ProviderPackage.Definition` and exposes `model(modelID, settings)`, where settings are serializable provider configuration plus common `headers`, `body`, and `limits` overlays.
 
 ```ts
-import { model } from "@opencode/ai/providers/openai/responses"
+import { model } from "@opencode-ai/ai/providers/openai/responses"
 
 const selected = model("gpt-5", {
   apiKey,

packages/ai/README.md

@@ -1,12 +1,12 @@
-# @opencode/ai
+# @opencode-ai/ai
 
-Schema-first language model and image-generation APIs built with Effect.
+Schema-first AI primitives for opencode. Provider quirks live in adapters, not in calling code.
 
 ```ts
 import { Effect, Layer } from "effect"
-import { LLM, LLMClient } from "@opencode/ai"
-import { RequestExecutor } from "@opencode/ai/route"
-import { OpenAI } from "@opencode/ai/providers"
+import { LLM, LLMClient } from "@opencode-ai/ai"
+import { RequestExecutor } from "@opencode-ai/ai/route"
+import { OpenAI } from "@opencode-ai/ai/providers"
 
 const model = OpenAI.configure({ apiKey: process.env.OPENAI_API_KEY }).responses("gpt-4o-mini")
 
@@ -29,123 +29,13 @@ await Effect.runPromise(program.pipe(Effect.provide(llmLayer)))
 
 Run `LLMClient.stream(request)` instead of `generate` when you want incremental `LLMEvent`s. The event stream is provider-neutral — same shape across OpenAI Chat, OpenAI Responses, Anthropic Messages, Gemini, Bedrock Converse, and any OpenAI-compatible deployment.
 
-## MiniMax
-
-MiniMax defaults to its Messages API and reads `MINIMAX_API_KEY` when `apiKey` is omitted:
-
-```ts
-import { Effect, Layer } from "effect"
-import { LLM, LLMClient } from "@opencode/ai"
-import { MiniMax } from "@opencode/ai/providers"
-import { RequestExecutor } from "@opencode/ai/route"
-
-const minimax = MiniMax.configure({ apiKey: process.env.MINIMAX_API_KEY })
-const request = LLM.request({
-  model: minimax.model("MiniMax-M3"), // also minimax.messages("MiniMax-M3")
-  prompt: "What is 173 multiplied by 219?",
-  providerOptions: { thinking: { type: "adaptive" } },
-  generation: { maxTokens: 1536 },
-})
-
-const layer = LLMClient.layer.pipe(Layer.provide(RequestExecutor.fetchLayer))
-const response = await Effect.runPromise(LLMClient.generate(request).pipe(Effect.provide(layer)))
-console.log(response.text)
-```
-
-Select `minimax.chat("MiniMax-M3")` or `minimax.responses("MiniMax-M3")` for MiniMax's native Chat Completions
-and Responses APIs. The matching package entrypoints are `@opencode/ai/providers/minimax/messages`,
-`@opencode/ai/providers/minimax/chat`, and `@opencode/ai/providers/minimax/responses`.
-
-- **Messages:** M3 thinking defaults off. Set `thinking: { type: "adaptive" }` to enable it or
-  `thinking: { type: "disabled" }` to disable it.
-- **Chat:** M3 thinking defaults on and uses the same `thinking` control. The provider enables `reasoning_split`
-  by default so reasoning is separate from answer text; `reasoningSplit: false` selects native `<think>`-tagged text.
-- **Responses:** M3 reasoning defaults off. `reasoningEffort: "none"` disables it; `"minimal"`, `"low"`,
-  `"medium"`, and `"high"` enable reasoning without changing its depth.
-
-M2.x models always think, even when a disabling option is supplied. For tool continuations, retain the complete
-`response.message` in history before adding `Message.tool(...)` results; this preserves reasoning and any signatures.
-
-The default API bases are `https://api.minimax.io/anthropic/v1` for Messages and `https://api.minimax.io/v1` for
-Chat and Responses. `configure({ baseURL })` replaces the selected API's base, including its version prefix.
-
-## Meta
-
-Use Meta's direct [Model API](https://dev.meta.ai/docs/overview) with `META_API_KEY`:
-
-```ts
-import { Meta } from "@opencode/ai/providers"
-
-const meta = Meta.configure() // or Meta.configure({ apiKey })
-const request = LLM.request({
-  model: meta.responses("muse-spark-1.3"), // meta.model(...) also selects Responses
-  prompt: "What is 173 multiplied by 219? Reply with the integer.",
-  providerOptions: { reasoningEffort: "low" },
-  generation: { maxTokens: 1024 },
-})
-```
-
-`meta.chat("muse-spark-1.3")` selects Chat Completions; `meta.messages("muse-spark-1.3")` selects
-the Anthropic-compatible Messages API. All use `https://api.meta.ai/v1`. The package entrypoints
-`@opencode/ai/providers/meta/responses`, `meta/chat`, and `meta/messages` expose `model(modelID, settings)`.
-
-[Muse Spark](https://dev.meta.ai/docs/models) supports `minimal`, `low`, `medium`, `high`, and
-`xhigh` reasoning effort; standard-tier 1.3 also supports `max`. Omitting effort uses the model's
-default. Muse Spark always reasons and rejects `none`. The output-token budget includes private reasoning.
-
-Responses defaults to `store: false` and `include: ["reasoning.encrypted_content"]`. Preserve
-`response.message` along with matching `Message.tool(...)` results in subsequent requests to replay
-reasoning through tool loops. Optional `reasoningSummary: "auto"` requests a readable summary.
-For server-managed history, override `store: true, include: []` and send the response ID through
-`http: { body: { previous_response_id: responseID } }` with only the new input.
-Chat Completions redacts private reasoning and cannot carry it between calls.
-Responses and Chat support only `toolChoice: "auto"` (the default). Messages also accepts `"none"`;
-its documented forced `"any"` choice currently returns HTTP 400. Messages defaults to adaptive thinking
-with `display: "omitted"`, preserving encrypted `redacted_thinking` in `response.message`. Use
-`providerOptions: { effort: "low" }` for depth or `thinking: { type: "enabled", budgetTokens: 1024 }`
-for budget compatibility (with `generation.maxTokens > 1024`).
-
-Add `tools: [Meta.webSearch()]` to a Spark Responses or Messages request for hosted web search.
-Responses exposes hosted results and URL citations in text-part `providerMetadata.meta.annotations`.
-To include search result lists, set `include: ["reasoning.encrypted_content", "web_search_call.results"]`.
-Messages exposes hosted search calls; the recorded Messages API stream does not supply structured
-citations or separate result blocks. Retain `response.message` for either API's continuation.
-
-Use `Image.generate` for one-off generation or editing:
-
-```ts
-import { Image, ImageInput } from "@opencode/ai"
-
-const generation = Image.generate({
-  model: meta.image("muse-image-1.0"),
-  prompt: "A flat black square on a white background.",
-  options: { n: 1, reasoningStrength: "low" },
-})
-
-const edit = Image.generate({
-  model: meta.image("muse-image-1.0"),
-  prompt: "Make the square purple.",
-  images: [ImageInput.bytes(imageBytes, "image/webp")],
-  options: { outputFormat: "png", reasoningStrength: "low" },
-})
-```
-
-The default image format is WEBP; `outputFormat` also accepts PNG/JPEG and `responseFormat: "url"`
-returns a signed URL. `size` is an aspect-ratio hint. For conversational images, select
-`meta.responses("muse-image-1.0")` with `tools: [Meta.imageGeneration({ reasoningStrength: "low" })]`.
-Generated images are provider-executed tool results with file content. Retain `response.message` to
-replay the signed image handle on the next request. Muse Image accepts only the `image_generation` tool.
-
-Meta Responses is explicitly HTTP/SSE-only and does not use WebSockets, even when a caller supplies
-`StreamOptions.webSocket`. The public `/v1/responses` endpoint rejects WebSocket upgrades with HTTP 405 (`Allow: POST`).
-
 ## Image generation
 
 Use `Image.generate` with an image model for direct asset generation:
 
 ```ts
-import { Image, ImageInput } from "@opencode/ai"
-import { OpenAI } from "@opencode/ai/providers"
+import { Image, ImageInput } from "@opencode-ai/ai"
+import { OpenAI } from "@opencode-ai/ai/providers"
 
 const program = Effect.gen(function* () {
   const response = yield* Image.generate({
@@ -241,7 +131,7 @@ yield *
 Google's current Gemini image models use the same direct API:
 
 ```ts
-import { Google } from "@opencode/ai/providers"
+import { Google } from "@opencode-ai/ai/providers"
 
 const googleProgram = Effect.gen(function* () {
   const response = yield* Image.generate({
@@ -317,183 +207,29 @@ The hosted result is represented as a provider-executed tool call and tool resul
 
 ## Testing
 
-Use the deterministic test client from `@opencode/ai/testing` to script provider-neutral responses and inspect
+Use the deterministic test client from `@opencode-ai/ai/testing` to script provider-neutral responses and inspect
 the requests sent by code under test:
 
 ```ts
 import { Effect } from "effect"
-import { TestLLM } from "@opencode/ai/testing"
+import { TestLLM } from "@opencode-ai/ai/testing"
 
+const testLLM = TestLLM.layer({
+  fallback: TestLLM.text("Hello from the test model", "text-1"),
+})
+
+// TestLLM.clientLayer provides LLMClient.Service and consumes TestLLM.Service.
 const programWithTestClient = Effect.gen(function* () {
-  const test = yield* TestLLM.Test
-  yield* test.push(TestLLM.text("Hello from the test model", "text-1"))
   const result = yield* program
-  console.log(yield* test.requests())
+  const test = yield* TestLLM.Service
+  console.log(test.requests)
   return result
-}).pipe(Effect.provide(TestLLM.testLayer()))
+}).pipe(Effect.provide(TestLLM.clientLayer), Effect.provide(testLLM))
 ```
 
-`testLayer()` provides the same object under `LLMClient.Service` and `TestLLM.Test`. Production consumes the
-normal client; tests use the additional controls. Each layer build has fresh state.
-
-- `test.push(...)` queues one-shot responses in execution order. Each argument is one response.
-- `test.always(response)` installs a repeatable fallback. The layer's `fallback` option sets its initial value.
-- `test.serve(request => response)` installs a request-dependent fallback. `always` and `serve` replace each
-  other without changing queued replies; queued replies take precedence.
-- `test.requests()` returns an array snapshot. `transformRequest` changes only the recorded observation;
-  `serve` receives the original canonical request.
-- `test.wait(count)` waits for request arrivals, not output or completion, and supports concurrent waiters.
-- `test.gate()` returns a scoped gate with countable `started` notifications and a `release` Effect. Release
-  unblocks all requests captured by that gate; closing its scope also releases it. Effect-aware test runners
-  already provide Scope.
-
-Constructing `stream()` or `generate()` does not record a request, invoke a responder, or consume a script.
-Each execution does. An exhausted queue without a fallback defects immediately rather than waiting for a
-future reply.
-
-Generation responses remain canonical event arrays or arbitrary `Stream<LLMEvent, AIError>` values. The client consumes
-supplied streams directly, preserving failure identity, finalizers, incomplete output, and post-finish tails;
-it does not repair or truncate them.
-
-For explicit compaction, script a `CompactionResponse` through `push`, `always`, or `serve`. Its `replacement` contains the next context window, including retained user messages. The client returns that result and usage directly, with the same lazy request recording and gates. Generation and compaction reject fixtures for the wrong operation instead of converting between response shapes.
-
-For `compact(request, { mechanism: "trigger" })`, script a `CompactionCheckpointResponse` instead. It carries `checkpoint`, `responseID`, and optional `usage`. Endpoint and trigger calls reject each other's fixtures; both share the same queue, gates, lazy recording, and fallback controls.
-
-The published legacy `Service`, `layer`, `clientLayer`, and module-level controls remain available as adapters
-over the same implementation, including the legacy live `requests` array. New tests should use `Test` and
-`testLayer`.
-
-## Provider compaction
-
-Compaction is opt-in. The package supports automatic compaction in OpenAI/Azure Responses and Anthropic Messages (including Claude on Vertex), and explicit compaction calls in OpenAI/Azure/xAI Responses. Model and deployment support still depends on the provider.
-
-This is different from prompt caching, server-side history storage, or truncation. Compaction returns provider-owned context that must be replayed to continue the conversation.
-
-### Explicit compaction
-
-`LLMClient.compact(request)` (equivalently, `{ mechanism: "endpoint" }`) is the caller-controlled operation for OpenAI, Azure, and xAI Responses. It performs exactly one HTTP call to `/responses/compact`, using the selected route's endpoint, credentials, query, and HTTP middleware. It returns a `CompactionResponse` with `replacement: Message[]` and optional `usage`, not a normal generation response. This mechanism does not accept a WebSocket executor.
-
-Prefer this operation, where supported, when the application owns compaction policy and durable context updates.
-
-```ts
-const result = yield * LLMClient.compact(request)
-const next = LLMRequest.update(request, {
-  messages: result.replacement,
-})
-const response = yield * LLMClient.generate(next)
-```
-
-`replacement` replaces the complete input window. Do not append it to the original transcript or extract only the encrypted item: the provider may retain additional messages in its output. Retained user and assistant messages remain ordinary messages with typed text, media, or reasoning parts, in their original order. Provider-specific message IDs, status, and phase use `providerMetadata`, not a raw output array hidden in an assistant message. Unsupported returned item types fail explicitly.
-
-The selected model carries explicit-compaction capability through request construction and updates. Calls using unsupported routes fail type checking. When the model is selected dynamically, narrow the request with `LLMClient.canCompact(request)` before calling `LLMClient.compact`; a model or route switch does not inherit the old capability. Runtime validation still rejects unsupported calls from untyped consumers. Capability describes the route's API, not whether every model or custom deployment supports the operation.
-
-Generation-only body overlays such as `stream` and `store` are not sent to the compact endpoint. Supported compact controls such as service tier and prompt-cache settings preserve request defaults and HTTP-overlay precedence. Retained image and file detail settings survive serialization and replay.
-
-The input must still fit the model's context window. Explicit compaction is not an overflow-recovery operation. Anthropic does not expose this operation in this package; its in-band compaction remains available below. Compatible routes do not inherit an explicit compact endpoint simply because they use a Responses protocol.
-
-### Streamed checkpoint compaction
-
-OpenAI Responses also exposes a separate, explicitly selected mechanism:
-
-```ts
-const result =
-  yield *
-  LLMClient.compact(request, {
-    mechanism: "trigger",
-    webSocket, // Optional: without it, the request uses HTTP/SSE.
-  })
-
-result.checkpoint // Successful encrypted CompactionPart.
-result.responseID
-result.usage
-```
-
-This appends a native `compaction_trigger` control item to the full input and sends a normal Responses request, with tools and instructions retained, `stream: true`, `store: false`, and parallel tool calls enabled. It removes normal-answer text/output-format controls, forced tool choices, output-token/tool-call limits, and automatic `context_management`. Body overlays cannot replace `input` or supply `previous_response_id`/`conversation`; the complete canonical history is required for safe stateless replay. Request metadata, auth, headers, query parameters, service tier, and supported prompt-cache settings are preserved.
-
-Only a successful `response.completed` with a response ID and exactly one logical encrypted checkpoint succeeds. Repeated item events are correlated by ID/output slot, including ID-less checkpoints. Other output is ignored, not returned as assistant text or dispatched as tools. Failed, incomplete, malformed, and interrupted responses return errors rather than partial checkpoints.
-
-The result is **not a replacement window**. The caller selects retained history, combines it with `result.checkpoint`, and durably installs it before continuing. The operation does not choose a retention budget, prune messages, or modify the original request.
-
-The supplied WebSocket executor can reuse a compatible append baseline for the compaction request. On completion the protocol supplies no continuation checkpoint, clearing the old baseline so the next generation sends the newly installed window in full. Validation occurs before transport completion is acknowledged. There is no operation-level retry or fallback to `/responses/compact`; existing safe transport fallback may use SSE, with full history and no connection-local response ID.
-
-Trigger support is separate from endpoint support. Only the OpenAI Responses route advertises it; Azure, xAI, Chat, and compatible Responses routes do not inherit it. Untyped calls still fail before sending: missing route capabilities return `UnsupportedOperation`, while unknown mechanism names and invalid inputs return `InvalidRequest`. Dynamic callers must narrow for the selected mechanism:
-
-```ts
-if (LLMClient.canCompact(request, { mechanism: "trigger" })) {
-  const result = yield * LLMClient.compact(request, { mechanism: "trigger" })
-}
-```
-
-This capability describes protocol implementation, **not universal availability on OpenAI API deployments**. The host application owns subscription/deployment eligibility, OAuth, endpoint selection, and deployment-specific headers. Local protocol/socket tests do not establish live provider support.
-
-### Advanced: in-band compaction
-
-`providerOptions.contextManagement` lets the provider decide when to compact during an ordinary `generate` or `stream` call. This is an advanced option for callers that own persistence and recovery: persist the complete assistant message, including its checkpoint, before continuing. Enabling the option does not provide durable checkpoint storage, interruption recovery, or model-switch policy. Keep the prior context until a successful checkpoint has been persisted.
-
-Inside an `Effect.gen`, enable OpenAI compaction with typed provider options:
-
-```ts
-import { LLM, LLMClient, LLMRequest, Message } from "@opencode/ai"
-import { OpenAI } from "@opencode/ai/providers"
-
-const request = LLM.request({
-  model: OpenAI.configure({ apiKey }).responses("gpt-5.3-codex"),
-  messages,
-  providerOptions: {
-    contextManagement: [{ type: "compaction", compactThreshold: 200_000 }],
-  },
-})
-const response = yield * LLMClient.generate(request)
-const next = LLMRequest.update(request, {
-  messages: [...request.messages, response.message, Message.user("Continue")],
-})
-```
-
-`store: false` remains the default. Keep the entire `response.message`, not just `response.text`. Compaction events become ordered `CompactionPart`s alongside text and reasoning. The conversation contains everything needed to continue; there is no separate replay object or hidden provider transcript.
-
-A compaction part has `provider` and exactly one representation: `encrypted` for Responses, or `text` for Anthropic. Responses also preserves the optional checkpoint `id`. These fields survive message serialization without becoming visible assistant text. Sending a checkpoint to another provider or an incompatible API fails rather than silently losing context.
-
-```ts
-import { CompactionPart, ProviderID } from "@opencode/ai"
-
-CompactionPart.make({ provider: ProviderID.make("openai"), id: "cmp_123", encrypted: "..." })
-CompactionPart.make({ provider: ProviderID.make("anthropic"), text: "Summary of the conversation..." })
-```
-
-For Anthropic, use:
-
-```ts
-providerOptions: {
-  contextManagement: {
-    edits: [{
-      type: "compact_20260112",
-      trigger: { type: "input_tokens", value: 150_000 },
-      pauseAfterCompaction: true,
-      instructions: "Summarize the task and decisions. Do not call tools while summarizing.",
-    }],
-  },
-}
-```
-
-- The trigger is optional (provider default: 150,000 tokens), with a minimum of 50,000.
-- Custom instructions replace Anthropic's default summarization instructions.
-- The route adds `compact-2026-01-12` to existing beta headers, including when replaying a checkpoint without enabling new compactions.
-- A pause is exposed as `response.finishReason.raw === "compaction"`. It occurs only if the threshold triggers compaction: `pauseAfterCompaction` does not mean "compact now". The caller explicitly issues the next request; the package never automatically resumes.
-- Anthropic can return a compaction block with `content: null` when summarization fails. This becomes a compaction part with `text: null`, which is **not** a successful replacement for prior history. The package never prunes history automatically.
-- `Usage` totals include all reported Anthropic `usage.iterations`, including compaction. `contextTokens` separately reports the final message iteration's inclusive input size, when available. A compaction-only pause does not report a post-compaction context size. Raw iteration usage remains in `providerMetadata`.
-
-### Recording tests
-
-Tests cover serialized round trips, real local HTTP plus a tool loop, WebSocket recovery, provider errors, malformed blocks, and usage accounting. Live provider tests are gated by `RECORD=true` and the relevant API keys:
-
-```sh
-# Run from packages/ai. Only records the selected new cassette group.
-RECORD=true RECORDED_PREFIX=openai-compaction bun test test/provider/compaction.recorded.test.ts
-RECORD=true RECORDED_PREFIX=xai-compaction bun test test/provider/compaction.recorded.test.ts
-RECORD=true RECORDED_PREFIX=anthropic-compaction bun test test/provider/compaction.recorded.test.ts
-```
-
-Provider references: [OpenAI](https://developers.openai.com/api/docs/guides/compaction), [Azure](https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/responses#server-side-compaction), [Anthropic](https://platform.claude.com/docs/en/build-with-claude/compaction), [xAI](https://docs.x.ai/developers/advanced-api-usage/context-compaction).
+`TestLLM.push(...)` scripts one-shot responses, `TestLLM.always(...)` changes the fallback, and
+`TestLLM.wait(...)` lets concurrent tests wait until a request has arrived. Every received canonical request is
+available on the yielded `TestLLM.Service`.
 
 ## Caching
 
@@ -501,7 +237,7 @@ Prompt caching is **on by default**. Every `LLMRequest` resolves to `cache: "aut
 
 ### Auto placement
 
-`"auto"` places up to four breakpoints — the last tool definition, the first system part, the last system part when distinct, and the final message boundary. These expose successively larger reusable prefixes for tool definitions, system instructions, and the active conversation. The rolling final-message boundary advances on every request so recent conversation prefixes remain reusable during tool loops.
+`"auto"` places up to four breakpoints — the last tool definition, the first system part, the last system part when distinct, and the final message boundary. These expose successively larger reusable prefixes for tools, the base agent, project instructions, and the active conversation. The rolling final-message boundary advances on every request so recent conversation prefixes remain reusable during tool loops.
 
 Tools precede every system and conversation block in the provider prefix, so tool definitions must remain byte-stable and deterministically ordered for downstream breakpoints to remain reusable.
 
@@ -560,7 +296,7 @@ Normalized cache usage is read back into `response.usage.cacheReadInputTokens` a
 Provider facades configure endpoint/auth/deployment details first, then expose model selectors that take only a model or deployment id. The selected model carries the executable route value used at runtime.
 
 ```ts
-import { OpenAI, CloudflareAIGateway } from "@opencode/ai/providers"
+import { OpenAI, CloudflareAIGateway } from "@opencode-ai/ai/providers"
 
 const openai = OpenAI.configure({ apiKey: process.env.OPENAI_API_KEY }).responses("gpt-4o-mini")
 const gateway = CloudflareAIGateway.configure({
@@ -569,42 +305,31 @@ const gateway = CloudflareAIGateway.configure({
 }).model("workers-ai/@cf/meta/llama-3.1-8b-instruct")
 ```
 
-Included LLM providers: OpenAI, Anthropic, Google (Gemini), Google Vertex, Amazon Bedrock, Azure OpenAI, Baseten, Cerebras, Cloudflare AI Gateway, Cloudflare Workers AI, DeepInfra, DeepSeek, Fireworks, Groq, Mistral, OpenRouter, TogetherAI, and xAI. Z.ai currently exposes image generation. Generic Chat Completions, Responses, and Anthropic Messages-compatible entrypoints support custom endpoints.
+Included providers: OpenAI, Anthropic, Google (Gemini), Google Vertex Gemini and Anthropic, Amazon Bedrock, Azure OpenAI, Cloudflare AI Gateway, Cloudflare Workers AI, GitHub Copilot, OpenRouter, xAI, Z.ai, plus generic OpenAI-compatible Chat and Responses entrypoints and an Anthropic Messages-compatible entrypoint.
 
-Each named provider owns its module, endpoint, authentication, and route setup. Providers with the same wire format compose the shared protocol directly:
+### Package-like entrypoints
+
+Native catalog integrations load provider behavior through package-like entrypoints. These are export paths from the same `@opencode-ai/ai` npm package, not independently published packages. Each entrypoint exports the same `model(modelID, settings)` contract, and `settings` contains serializable provider configuration plus common `headers` and `body` overlays.
 
 ```ts
-import { DeepSeek, Fireworks } from "@opencode/ai/providers"
-
-const deepseek = DeepSeek.configure({ apiKey }).model("deepseek-chat")
-const fireworks = Fireworks.configure({ apiKey }).model("accounts/fireworks/models/my-model")
-```
-
-The former `OpenAICompatible.baseten`, `.cerebras`, `.deepinfra`, `.deepseek`, `.fireworks`, `.groq`, and `.togetherai` presets are replaced by the top-level `Baseten`, `Cerebras`, `DeepInfra`, `DeepSeek`, `Fireworks`, `Groq`, and `TogetherAI` exports. Use `CloudflareAIGateway` and `CloudflareWorkersAI` directly; each has its own module. `OpenAICompatible` configures generic endpoints with an explicit `baseURL`.
-
-### Provider entrypoints
-
-Provider modules are available through dedicated exports from `@opencode/ai`. Each LLM entrypoint exports `model(modelID, settings)`, where `settings` contains provider configuration plus common `headers` and `body` overlays.
-
-```ts
-import { model } from "@opencode/ai/providers/openai/responses"
+import { model } from "@opencode-ai/ai/providers/openai/responses"
 
 const selected = model("gpt-5", {
   apiKey: process.env.OPENAI_API_KEY,
-  headers: { "x-application": "example" },
+  headers: { "x-application": "opencode" },
 })
 ```
 
-APIs have separate entrypoints:
+OpenAI Chat and OpenAI Responses are separate semantic entrypoints:
 
-- `@opencode/ai/providers/openai/chat`
-- `@opencode/ai/providers/openai/responses`
-- `@opencode/ai/providers/openai-compatible/responses`
-- `@opencode/ai/providers/anthropic-compatible`
-- `@opencode/ai/providers/google-vertex/gemini`
-- `@opencode/ai/providers/google-vertex/chat`
-- `@opencode/ai/providers/google-vertex/responses`
-- `@opencode/ai/providers/google-vertex/messages`
+- `@opencode-ai/ai/providers/openai/chat`
+- `@opencode-ai/ai/providers/openai/responses`
+- `@opencode-ai/ai/providers/openai-compatible/responses`
+- `@opencode-ai/ai/providers/anthropic-compatible`
+- `@opencode-ai/ai/providers/google-vertex/gemini`
+- `@opencode-ai/ai/providers/google-vertex/chat`
+- `@opencode-ai/ai/providers/google-vertex/responses`
+- `@opencode-ai/ai/providers/google-vertex/messages`
 
 OpenAI Responses has one semantic route and uses HTTP by default. Advanced callers may supply a per-call WebSocket channel executor through `StreamOptions`; transport policy does not change provider settings, model identity, or route identity. The provider-neutral Open Responses implementation owns the reusable WebSocket request and event contract, while each provider opts in with its own handshake and connection policy. Azure follows the same Chat/Responses split at `providers/azure/chat` and `providers/azure/responses`. Generic OpenAI-compatible Chat remains at `providers/openai-compatible`; the Responses adapter at `providers/openai-compatible/responses` uses the provider-neutral Open Responses protocol. OpenAI Responses extends that baseline with OpenAI tools, event variants, metadata, and defaults. Generic Anthropic Messages-compatible providers use `providers/anthropic-compatible`, which the named Anthropic provider composes. Google Gemini and Amazon Bedrock expose their single native API through their existing provider paths.
 
@@ -613,36 +338,32 @@ Vertex Gemini, Vertex Chat, Vertex Responses, and Vertex Messages are separate A
 Tuned Vertex Gemini deployments use model ids shaped like `endpoints/1234567890` and require OAuth or ADC; Vertex express-mode API keys support publisher models only.
 
 ```ts
-import { model } from "@opencode/ai/providers/google-vertex/gemini"
+import { model } from "@opencode-ai/ai/providers/google-vertex/gemini"
 
 model("gemini-3.5-flash", { project: "my-project", location: "global" })
 ```
 
 ```ts
-import { model } from "@opencode/ai/providers/google-vertex/chat"
+import { model } from "@opencode-ai/ai/providers/google-vertex/chat"
 
 model("deepseek-ai/deepseek-v3.2-maas", { project: "my-project", location: "global" })
 ```
 
 ```ts
-import { model } from "@opencode/ai/providers/google-vertex/responses"
+import { model } from "@opencode-ai/ai/providers/google-vertex/responses"
 
 model("xai/grok-4.20-reasoning", { project: "my-project", location: "global" })
 ```
 
 ```ts
-import { model } from "@opencode/ai/providers/google-vertex/messages"
+import { model } from "@opencode-ai/ai/providers/google-vertex/messages"
 
 model("claude-sonnet-4-6", { project: "my-project", location: "global" })
 ```
 
-Additional provider entrypoints include:
+Provider facades such as `OpenAI.configure(...).responses(...)` remain the direct application API. Package-like entrypoints are the self-similar loading contract used when a catalog selects behavior by export path.
 
-- `@opencode/ai/providers/baseten`
-- `@opencode/ai/providers/deepseek`
-- `@opencode/ai/providers/fireworks`
-- `@opencode/ai/providers/cloudflare-ai-gateway`
-- `@opencode/ai/providers/cloudflare-workers-ai`
+Other provider exports listed above remain direct facades until they explicitly implement the package-like contract. Exporting a provider facade does not implicitly make it a catalog-loadable provider package.
 
 ## Provider options & HTTP overlays
 
@@ -669,7 +390,7 @@ LLM.request({
 
 ## Routes
 
-Compose a route with `Route.make({ protocol, endpoint, auth, framing, ... })`. The route owns endpoint/auth/framing and the protocol owns body construction plus stream parsing. Transports receive the route's endpoint and auth when preparing requests. Unsupported request shapes fail during protocol lowering.
+Adding a new model or deployment is usually 5-15 lines using `Route.make({ protocol, endpoint, auth, framing, ... })`. The route owns endpoint/auth/framing and the protocol owns body construction plus stream parsing. Transports are reusable IO templates that receive route endpoint/auth at compile time. Capability/catalog metadata lives outside this low-level package; unsupported request shapes fail during protocol lowering. See `AGENTS.md` for the architectural detail.
 
 ## Effect
 

packages/ai/example/tutorial.ts

@@ -1,7 +1,7 @@
 import { Config, Effect, Formatter, Layer, Schema, Stream } from "effect"
-import { LLM, LLMClient, LLMRequest, Message, ProviderID, Tool, ToolRuntime } from "@opencode/ai"
-import { Route, Auth, Endpoint, Framing, Protocol, RequestExecutor } from "@opencode/ai/route"
-import { OpenAI } from "@opencode/ai/providers"
+import { LLM, LLMClient, LLMRequest, Message, ProviderID, Tool, ToolRuntime } from "@opencode-ai/ai"
+import { Route, Auth, Endpoint, Framing, Protocol, RequestExecutor } from "@opencode-ai/ai/route"
+import { OpenAI } from "@opencode-ai/ai/providers"
 
 /**
  * A runnable walkthrough of the LLM package use-site API.

packages/ai/package.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://json.schemastore.org/package.json",
   "version": "1.17.20",
-  "name": "@opencode/ai",
+  "name": "@opencode-ai/ai",
   "type": "module",
   "license": "MIT",
   "scripts": {
@@ -21,17 +21,16 @@
   "devDependencies": {
     "@clack/prompts": "1.0.0-alpha.1",
     "@effect/platform-node": "catalog:",
-    "@opencode/http-recorder": "workspace:*",
+    "@opencode-ai/http-recorder": "workspace:*",
     "@tsconfig/bun": "catalog:",
     "@types/bun": "catalog:",
     "@typescript/native-preview": "catalog:",
     "typescript": "catalog:"
   },
   "dependencies": {
-    "@aws-sdk/credential-providers": "3.1057.0",
     "@smithy/eventstream-codec": "4.2.14",
     "@smithy/util-utf8": "4.2.2",
-    "@opencode/schema": "workspace:*",
+    "@opencode-ai/schema": "workspace:*",
     "aws4fetch": "1.0.20",
     "effect": "catalog:",
     "google-auth-library": "10.5.0"

packages/ai/script/publish.ts

@@ -1,5 +1,5 @@
 #!/usr/bin/env bun
-import { Script } from "@opencode/script"
+import { Script } from "@opencode-ai/script"
 import { $ } from "bun"
 import { fileURLToPath } from "url"
 

packages/ai/script/setup-recording-env.ts

@@ -7,8 +7,7 @@ import { AwsV4Signer } from "aws4fetch"
 import { Config, ConfigProvider, Effect, FileSystem, PlatformError, Redacted } from "effect"
 import { FetchHttpClient, HttpClient, HttpClientRequest, type HttpClientResponse } from "effect/unstable/http"
 import * as ProviderShared from "../src/protocols/shared"
-import { CloudflareAIGateway } from "../src/providers/cloudflare-ai-gateway.js"
-import { CloudflareWorkersAI } from "../src/providers/cloudflare-workers-ai.js"
+import * as Cloudflare from "../src/providers/cloudflare"
 
 type Provider = {
   readonly id: string
@@ -121,11 +120,11 @@ const PROVIDERS: ReadonlyArray<Provider> = [
     ],
     validate: (env) =>
       validateChat({
-        url: `${CloudflareAIGateway.baseURL({
+        url: `${Cloudflare.aiGatewayBaseURL({
           accountId: env.CLOUDFLARE_ACCOUNT_ID,
           gatewayId: env.CLOUDFLARE_GATEWAY_ID || undefined,
         })}/chat/completions`,
-        token: Redacted.make(envValue(env, CloudflareAIGateway.authEnvVars)),
+        token: Redacted.make(envValue(env, Cloudflare.aiGatewayAuthEnvVars)),
         tokenHeader: "cf-aig-authorization",
         model: "workers-ai/@cf/meta/llama-3.1-8b-instruct",
       }),
@@ -141,8 +140,8 @@ const PROVIDERS: ReadonlyArray<Provider> = [
     ],
     validate: (env) =>
       validateChat({
-        url: `${CloudflareWorkersAI.baseURL({ accountId: env.CLOUDFLARE_ACCOUNT_ID })}/chat/completions`,
-        token: Redacted.make(envValue(env, CloudflareWorkersAI.authEnvVars)),
+        url: `${Cloudflare.workersAIBaseURL({ accountId: env.CLOUDFLARE_ACCOUNT_ID })}/chat/completions`,
+        token: Redacted.make(envValue(env, Cloudflare.workersAIAuthEnvVars)),
         model: "@cf/meta/llama-3.1-8b-instruct",
       }),
   },

packages/ai/src/cache-policy.ts

@@ -11,7 +11,7 @@
 // Manual `cache: CacheHint` placements on individual parts are preserved and
 // count against the four-breakpoint budget; auto only fills remaining slots.
 import { CacheHint, type CachePolicy, type CachePolicyObject } from "./schema/options.js"
-import { LLMRequest, Message, ToolDefinition, type ContentPart, type ToolEntry } from "./schema/messages.js"
+import { LLMRequest, Message, ToolDefinition, type ContentPart } from "./schema/messages.js"
 
 const AUTO: CachePolicyObject = {
   tools: true,
@@ -50,24 +50,18 @@ interface Budget {
   remaining: number
 }
 
-const markLastTool = (tools: ReadonlyArray<ToolEntry>, hint: CacheHint, budget: Budget): ReadonlyArray<ToolEntry> => {
-  const target = tools.at(-1)
-  if (target === undefined) return tools
-  if (target.type === "namespace") {
-    const nested = markLastTool(target.tools, hint, budget)
-    return nested === target.tools ? tools : [...tools.slice(0, -1), { ...target, tools: nested }]
-  }
-  if (target.cache || budget.remaining === 0) return tools
+const markLastTool = (
+  tools: ReadonlyArray<ToolDefinition>,
+  hint: CacheHint,
+  budget: Budget,
+): ReadonlyArray<ToolDefinition> => {
+  if (tools.length === 0) return tools
+  const last = tools.length - 1
+  if (tools[last]!.cache || budget.remaining === 0) return tools
   budget.remaining -= 1
-  return [...tools.slice(0, -1), new ToolDefinition({ ...target, cache: hint })]
+  return tools.map((tool, i) => (i === last ? new ToolDefinition({ ...tool, cache: hint }) : tool))
 }
 
-const countToolHints = (tools: ReadonlyArray<ToolEntry>): number =>
-  tools.reduce(
-    (count, tool) => count + (tool.type === "tool" ? (tool.cache === undefined ? 0 : 1) : countToolHints(tool.tools)),
-    0,
-  )
-
 const markSystemBoundaries = (system: LLMRequest["system"], hint: CacheHint, budget: Budget): LLMRequest["system"] => {
   if (system.length === 0) return system
   let changed = false
@@ -128,7 +122,7 @@ const markMessages = (
 }
 
 const countHints = (request: LLMRequest) =>
-  countToolHints(request.tools) +
+  request.tools.reduce((count, tool) => count + (tool.cache === undefined ? 0 : 1), 0) +
   request.system.reduce((count, part) => count + (part.cache === undefined ? 0 : 1), 0) +
   request.messages.reduce(
     (count, message) =>

packages/ai/src/llm.ts

@@ -12,10 +12,9 @@ import {
   LanguageModel,
   SystemPart,
   ToolChoice,
-  ToolEntry,
+  ToolDefinition,
   type ContentPart,
   type LanguageModelProviderOptions,
-  type ToolEntryInput,
 } from "./schema/index.js"
 import { make as makeTool, toDefinitions, type ToolSchema } from "./tool.js"
 
@@ -28,7 +27,7 @@ export type RequestInput<SelectedLanguageModel extends LanguageModel = LanguageM
   readonly system?: string | SystemPart | ReadonlyArray<SystemPart>
   readonly prompt?: string | ContentPart | ReadonlyArray<ContentPart>
   readonly messages?: ReadonlyArray<Message | Message.Input>
-  readonly tools?: ReadonlyArray<ToolEntryInput>
+  readonly tools?: ReadonlyArray<ToolDefinition.Input>
   readonly toolChoice?: ToolChoice.Input
   readonly generation?: GenerationOptions.Input
   readonly providerOptions?: NoInfer<LanguageModelProviderOptions<SelectedLanguageModel>>
@@ -57,7 +56,7 @@ export const request = <const SelectedLanguageModel extends LanguageModel>(
     ...rest,
     system: SystemPart.content(requestSystem),
     messages: [...(messages?.map(Message.make) ?? []), ...(prompt === undefined ? [] : [Message.user(prompt)])],
-    tools: tools?.map(ToolEntry.make) ?? [],
+    tools: tools?.map(ToolDefinition.make) ?? [],
     toolChoice: requestToolChoice ? ToolChoice.make(requestToolChoice) : undefined,
     generation: requestGeneration === undefined ? undefined : GenerationOptions.make(requestGeneration),
     providerOptions: requestProviderOptions,

packages/ai/src/protocols/anthropic-messages.ts

@@ -1,17 +1,13 @@
 import { Buffer } from "node:buffer"
 import { Effect, Option, Schema } from "effect"
-import { Tool } from "@opencode/schema/tool"
+import { Tool } from "@opencode-ai/schema/tool"
 import { Route } from "../route/client.js"
 import { Auth } from "../route/auth.js"
 import { Endpoint } from "../route/endpoint.js"
 import { Framing } from "../route/framing.js"
 import { Protocol } from "../route/protocol.js"
-import { Headers } from "effect/unstable/http"
-import { HttpTransport } from "../route/transport/index.js"
 import {
   AIError,
-  HttpOptions,
-  LLMRequest,
   LLMEvent,
   mergeJsonRecords,
   Usage,
@@ -19,6 +15,7 @@ import {
   type FinishReasonDetails,
   type FinishReason,
   type JsonSchema,
+  type LLMRequest,
   type MediaPart,
   type ProviderMetadata,
   type ToolCallPart,
@@ -64,8 +61,6 @@ export type ThinkingInput =
     ))
 
 export interface OptionsInput {
-  /** Advanced in-band compaction. The caller owns checkpoint persistence and recovery. */
-  readonly contextManagement?: ContextManagement
   readonly [key: string]: unknown
   readonly thinking?: ThinkingInput
   readonly effort?: string
@@ -94,23 +89,6 @@ export interface OptionsInput {
 
 export type ProviderOptionsInput = OptionsInput
 
-export const ContextManagement = Schema.Struct({
-  edits: Schema.Array(
-    Schema.Struct({
-      type: Schema.Literal("compact_20260112"),
-      trigger: Schema.optional(
-        Schema.Struct({
-          type: Schema.Literal("input_tokens"),
-          value: Schema.Int.check(Schema.isGreaterThanOrEqualTo(50000)),
-        }),
-      ),
-      pauseAfterCompaction: Schema.optional(Schema.Boolean),
-      instructions: Schema.optional(Schema.String),
-    }),
-  ),
-})
-export type ContextManagement = typeof ContextManagement.Type
-
 // =============================================================================
 // Request Body Schema
 // =============================================================================
@@ -258,12 +236,7 @@ const AnthropicUserBlock = Schema.Union([
   AnthropicToolResultBlock,
 ])
 type AnthropicUserBlock = Schema.Schema.Type<typeof AnthropicUserBlock>
-const AnthropicCompactionBlock = Schema.Struct({
-  type: Schema.Literal("compaction"),
-  content: Schema.NullOr(Schema.String),
-})
 const AnthropicAssistantBlock = Schema.Union([
-  AnthropicCompactionBlock,
   AnthropicTextBlock,
   AnthropicThinkingBlock,
   AnthropicRedactedThinkingBlock,
@@ -339,18 +312,6 @@ const AnthropicContainer = Schema.Union([
 ])
 
 const AnthropicBodyFields = {
-  context_management: Schema.optional(
-    Schema.Struct({
-      edits: Schema.Array(
-        Schema.Struct({
-          type: Schema.Literal("compact_20260112"),
-          trigger: ContextManagement.fields.edits.value.fields.trigger,
-          pause_after_compaction: Schema.optional(Schema.Boolean),
-          instructions: Schema.optional(Schema.String),
-        }),
-      ),
-    }),
-  ),
   model: Schema.String,
   system: optionalArray(AnthropicTextBlock),
   messages: Schema.Array(AnthropicMessage),
@@ -374,7 +335,7 @@ const AnthropicBodyFields = {
 export const AnthropicMessagesBody = Schema.Struct(AnthropicBodyFields)
 export type AnthropicMessagesBody = Schema.Schema.Type<typeof AnthropicMessagesBody>
 
-const AnthropicIterationUsage = Schema.StructWithRest(
+const AnthropicUsage = Schema.StructWithRest(
   Schema.Struct({
     input_tokens: optionalNull(Schema.Number),
     output_tokens: Schema.optional(Schema.Number),
@@ -393,13 +354,6 @@ const AnthropicIterationUsage = Schema.StructWithRest(
   }),
   [Schema.Record(Schema.String, Schema.Unknown)],
 )
-const AnthropicUsage = Schema.StructWithRest(
-  Schema.Struct({
-    ...AnthropicIterationUsage.schema.fields,
-    iterations: Schema.optional(Schema.Array(AnthropicIterationUsage)),
-  }),
-  [JsonObject],
-)
 type AnthropicUsage = Schema.Schema.Type<typeof AnthropicUsage>
 
 const AnthropicStreamBlock = Schema.Struct({
@@ -423,7 +377,6 @@ type AnthropicStreamBlock = Schema.Schema.Type<typeof AnthropicStreamBlock>
 const decodeAnthropicStreamBlock = Schema.decodeUnknownOption(AnthropicStreamBlock)
 
 const AnthropicStreamDelta = Schema.Struct({
-  content: optionalNull(Schema.String),
   type: Schema.optional(Schema.String),
   text: Schema.optional(Schema.String),
   thinking: Schema.optional(Schema.String),
@@ -453,8 +406,6 @@ const AnthropicEvent = Schema.Struct({
 type AnthropicEvent = Schema.Schema.Type<typeof AnthropicEvent>
 
 interface ParserState {
-  readonly provider: LLMRequest["model"]["provider"]
-  readonly compactions: Readonly<Record<number, string | null>>
   readonly providerMetadataKey: string
   readonly tools: ToolStream.State<number>
   readonly reasoningSignatures: Readonly<Record<number, string>>
@@ -880,7 +831,6 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
       const content: AnthropicUserBlock[] = []
       for (const part of message.content) {
         if (part.type === "text") {
-          if (part.text.trim().length === 0) continue
           content.push({ type: "text", text: part.text, cache_control: cacheControl(breakpoints, part.cache) })
           continue
         }
@@ -890,21 +840,14 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
         }
         return yield* ProviderShared.unsupportedContent("Anthropic Messages", "user", ["text", "media"])
       }
-      if (content.length > 0) messages.push({ role: "user", content })
+      messages.push({ role: "user", content })
       continue
     }
 
     if (message.role === "assistant") {
       const content: AnthropicAssistantBlock[] = []
       for (const part of message.content) {
-        if (part.type === "compaction") {
-          if (part.provider !== request.model.provider || part.text === undefined)
-            return yield* invalid("Compaction state must be replayed to its originating provider and API")
-          content.push({ type: "compaction", content: part.text })
-          continue
-        }
         if (part.type === "text") {
-          if (part.text.trim().length === 0) continue
           content.push({ type: "text", text: part.text, cache_control: cacheControl(breakpoints, part.cache) })
           continue
         }
@@ -948,7 +891,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
           `Anthropic Messages assistant messages only support text, reasoning, and tool-call content for now`,
         )
       }
-      if (content.length > 0) messages.push({ role: "assistant", content })
+      messages.push({ role: "assistant", content })
       continue
     }
 
@@ -1058,20 +1001,16 @@ const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function*
 })
 
 const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (request: LLMRequest) {
-  const management = yield* ProviderShared.validateWith(
-    Schema.decodeUnknownEffect(Schema.UndefinedOr(ContextManagement)),
-  )(request.providerOptions?.contextManagement)
   const generation = request.generation
   const toolSchemaCompatibility = request.model.compatibility?.toolSchema
   // Allocate the 4-breakpoint budget in invalidation order: tools → system →
   // messages. Tools live highest in the cache hierarchy, so when callers
   // over-mark we keep their tool hints and shed the message-tail ones first.
   const breakpoints = Cache.newBreakpoints(ANTHROPIC_BREAKPOINT_CAP)
-  const flattened = ProviderShared.flattenToolRequest(request)
   const tools =
-    flattened.tools.length === 0
+    request.tools.length === 0
       ? undefined
-      : flattened.tools.map((tool) =>
+      : request.tools.map((tool) =>
           lowerTool(
             breakpoints,
             tool,
@@ -1080,23 +1019,22 @@ const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (reques
         )
   // Anthropic rejects tool_choice when tools are absent; "none" is only meaningful with tools present.
   const toolChoice = tools === undefined || !request.toolChoice ? undefined : yield* lowerToolChoice(request.toolChoice)
-  const systemParts = request.system.filter((part) => part.text.length > 0)
   const system =
-    systemParts.length === 0
+    request.system.length === 0
       ? undefined
-      : systemParts.map((part) => ({
+      : request.system.map((part) => ({
           type: "text" as const,
           text: part.text,
           cache_control: cacheControl(breakpoints, part.cache),
         }))
-  const messages = yield* lowerMessages(flattened.request, breakpoints)
+  const messages = yield* lowerMessages(request, breakpoints)
   if (breakpoints.dropped > 0) {
     yield* Effect.logWarning(
       `Anthropic Messages: dropped ${breakpoints.dropped} cache breakpoint(s); the API allows at most ${ANTHROPIC_BREAKPOINT_CAP} per request.`,
     )
   }
   const options = yield* resolveOptions(request)
-  const body = {
+  return {
     model: request.model.id,
     system,
     messages,
@@ -1117,18 +1055,6 @@ const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (reques
     metadata: options.metadata,
     service_tier: options.service_tier,
   }
-  if (!management) return body
-  return {
-    ...body,
-    context_management: {
-      edits: management.edits.map((edit) => ({
-        type: edit.type,
-        trigger: edit.trigger,
-        pause_after_compaction: edit.pauseAfterCompaction,
-        instructions: edit.instructions,
-      })),
-    },
-  }
 })
 
 // =============================================================================
@@ -1150,31 +1076,18 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
 // expose that subset through `output_tokens_details.thinking_tokens`.
 const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string): Usage | undefined => {
   if (!usage) return undefined
-  const iterations = usage.iterations?.length ? usage.iterations : [usage]
-  const last = usage.iterations?.at(-1)
-  const nonCached = ProviderShared.sumTokens(...iterations.map((item) => item.input_tokens ?? undefined))
-  const cacheRead = ProviderShared.sumTokens(...iterations.map((item) => item.cache_read_input_tokens ?? undefined))
-  const cacheWrite = ProviderShared.sumTokens(
-    ...iterations.map((item) => item.cache_creation_input_tokens ?? undefined),
-  )
+  const nonCached = usage.input_tokens ?? undefined
+  const cacheRead = usage.cache_read_input_tokens ?? undefined
+  const cacheWrite = usage.cache_creation_input_tokens ?? undefined
   const inputTokens = ProviderShared.sumTokens(nonCached, cacheRead, cacheWrite)
-  const outputTokens = ProviderShared.sumTokens(...iterations.map((item) => item.output_tokens))
   return new Usage({
     inputTokens,
-    outputTokens,
-    contextTokens:
-      last?.type === "message"
-        ? ProviderShared.sumTokens(
-            last.input_tokens ?? undefined,
-            last.cache_read_input_tokens ?? undefined,
-            last.cache_creation_input_tokens ?? undefined,
-          )
-        : undefined,
+    outputTokens: usage.output_tokens,
     nonCachedInputTokens: nonCached,
     cacheReadInputTokens: cacheRead,
     cacheWriteInputTokens: cacheWrite,
-    reasoningTokens: ProviderShared.sumTokens(...iterations.map((item) => item.output_tokens_details?.thinking_tokens)),
-    totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
+    reasoningTokens: usage.output_tokens_details?.thinking_tokens,
+    totalTokens: ProviderShared.totalTokens(inputTokens, usage.output_tokens, undefined),
     providerMetadata: { [providerMetadataKey]: usage },
   })
 }
@@ -1196,7 +1109,6 @@ const mergeUsage = (left: Usage | undefined, right: Usage | undefined, providerM
   return new Usage({
     inputTokens,
     outputTokens,
-    contextTokens: right.contextTokens ?? left.contextTokens,
     nonCachedInputTokens,
     cacheReadInputTokens,
     cacheWriteInputTokens,
@@ -1255,6 +1167,7 @@ const onContentBlockStart = (
   event: AnthropicEvent & { readonly content_block: AnthropicStreamBlock },
 ): StepResult => {
   const block = event.content_block
+  if (!block) return [state, NO_EVENTS]
 
   if (block.type === "tool_use" || block.type === "server_tool_use") {
     if (event.index === undefined || !block.id) return [state, NO_EVENTS]
@@ -1349,16 +1262,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
 ) {
   const delta = event.delta
 
-  if (delta.type === "compaction_delta") {
-    if (event.index === undefined || !(event.index in state.compactions) || delta.content === undefined)
-      return yield* ProviderShared.eventError(ADAPTER, "Compaction delta is missing its block or content")
-    return [
-      { ...state, compactions: { ...state.compactions, [event.index]: delta.content } },
-      NO_EVENTS,
-    ] satisfies StepResult
-  }
-
-  if (delta.type === "text_delta" && delta.text) {
+  if (delta?.type === "text_delta" && delta.text) {
     if (!state.lifecycle.text.has(`text-${event.index ?? 0}`)) return [state, NO_EVENTS] satisfies StepResult
     const events: LLMEvent[] = []
     return [
@@ -1367,7 +1271,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
     ] satisfies StepResult
   }
 
-  if (delta.type === "thinking_delta" && delta.thinking) {
+  if (delta?.type === "thinking_delta" && delta.thinking) {
     if (!state.lifecycle.reasoning.has(`reasoning-${event.index ?? 0}`)) return [state, NO_EVENTS] satisfies StepResult
     const events: LLMEvent[] = []
     return [
@@ -1379,7 +1283,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
     ] satisfies StepResult
   }
 
-  if (delta.type === "signature_delta" && delta.signature) {
+  if (delta?.type === "signature_delta" && delta.signature) {
     const index = event.index ?? 0
     if (!state.lifecycle.reasoning.has(`reasoning-${index}`)) return [state, NO_EVENTS] satisfies StepResult
     return [
@@ -1391,7 +1295,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
     ] satisfies StepResult
   }
 
-  if (delta.type === "input_json_delta" && event.index !== undefined) {
+  if (delta?.type === "input_json_delta" && event.index !== undefined) {
     if (!delta.partial_json) return [state, NO_EVENTS] satisfies StepResult
     if (!state.tools[event.index]) return [state, NO_EVENTS] satisfies StepResult
     const result = ToolStream.appendExisting(
@@ -1416,18 +1320,6 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
   event: AnthropicEvent,
 ) {
   if (event.index === undefined) return [state, NO_EVENTS] satisfies StepResult
-  if (event.index in state.compactions) {
-    const { [event.index]: content, ...compactions } = state.compactions
-    const events: LLMEvent[] = []
-    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
-    events.push(
-      LLMEvent.compaction({
-        provider: state.provider,
-        text: content,
-      }),
-    )
-    return [{ ...state, compactions, lifecycle }, events] satisfies StepResult
-  }
   const result = yield* ToolStream.finish(ADAPTER, state.tools, event.index)
   const events: LLMEvent[] = []
   const resultEvents = result.events ?? []
@@ -1451,51 +1343,31 @@ const onMessageDelta = (
   event: AnthropicEvent & { readonly delta?: AnthropicStreamDelta },
 ): StepResult => {
   const usage = mergeUsage(state.usage, mapUsage(event.usage, state.providerMetadataKey), state.providerMetadataKey)
-  const pendingFinish = (() => {
-    const stopReason = event.delta?.stop_reason
-    if (stopReason === null || stopReason === undefined) return state.pendingFinish
-
-    const stopSequence = event.delta?.stop_sequence
-    const finishMetadata =
-      stopSequence === null || stopSequence === undefined
-        ? state.pendingFinish?.providerMetadata
-        : providerMetadata(state.providerMetadataKey, { stopSequence })
-    return {
-      reason: {
-        normalized: mapFinishReason(stopReason),
-        raw: stopReason,
-      },
-      providerMetadata: finishMetadata,
-    }
-  })()
   return [
     {
       ...state,
       usage,
-      pendingFinish,
+      pendingFinish: {
+        reason: {
+          normalized: mapFinishReason(event.delta?.stop_reason),
+          raw: event.delta?.stop_reason ?? undefined,
+        },
+        providerMetadata:
+          event.delta?.stop_sequence === null || event.delta?.stop_sequence === undefined
+            ? undefined
+            : providerMetadata(state.providerMetadataKey, { stopSequence: event.delta.stop_sequence }),
+      },
     },
     NO_EVENTS,
   ]
 }
 
 const onMessageStop = Effect.fn("AnthropicMessages.onMessageStop")(function* (state: ParserState) {
-  if (Object.keys(state.compactions).length)
-    return yield* ProviderShared.eventError(ADAPTER, "Response ended with an incomplete compaction block")
   const result = yield* ToolStream.finishAll(ADAPTER, state.tools)
   const events: LLMEvent[] = []
   const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
   events.push(...result.events)
-  const closed = Object.entries(state.reasoningSignatures).reduce(
-    (current, [index, signature]) =>
-      Lifecycle.reasoningEnd(
-        current,
-        events,
-        `reasoning-${index}`,
-        providerMetadata(state.providerMetadataKey, { signature }),
-      ),
-    lifecycle,
-  )
-  const finished = Lifecycle.finish(closed, events, {
+  const finished = Lifecycle.finish(lifecycle, events, {
     reason: state.pendingFinish?.reason ?? {
       normalized: "unknown",
       raw: undefined,
@@ -1525,21 +1397,16 @@ const onError = (event: AnthropicEvent) => {
   )
 }
 
-const STREAM_BLOCK_TYPES = new Set([
-  "compaction",
-  "text",
-  "thinking",
-  "redacted_thinking",
-  "tool_use",
-  "server_tool_use",
-])
-const STREAM_DELTA_TYPES = new Set([
-  "compaction_delta",
-  "text_delta",
-  "thinking_delta",
-  "signature_delta",
-  "input_json_delta",
-])
+const isKnownStreamBlockType = (type: string) =>
+  type === "text" ||
+  type === "thinking" ||
+  type === "redacted_thinking" ||
+  type === "tool_use" ||
+  type === "server_tool_use" ||
+  isServerToolResultType(type)
+
+const isKnownStreamDeltaType = (type: string) =>
+  type === "text_delta" || type === "thinking_delta" || type === "signature_delta" || type === "input_json_delta"
 
 const invalidStreamEvent = (event: AnthropicEvent) =>
   Effect.fail(
@@ -1568,16 +1435,7 @@ const step = (state: ParserState, event: AnthropicEvent) => {
   if (event.type === "content_block_start") {
     if (!ProviderShared.isRecord(event.content_block) || typeof event.content_block.type !== "string")
       return invalidStreamEvent(event)
-    if (event.content_block.type === "compaction") {
-      const decoded = Schema.decodeUnknownOption(AnthropicCompactionBlock)(event.content_block)
-      if (event.index === undefined || Option.isNone(decoded)) return invalidStreamEvent(event)
-      return Effect.succeed<StepResult>([
-        { ...state, compactions: { ...state.compactions, [event.index]: decoded.value.content } },
-        NO_EVENTS,
-      ])
-    }
-    if (!STREAM_BLOCK_TYPES.has(event.content_block.type) && !isServerToolResultType(event.content_block.type))
-      return Effect.succeed<StepResult>([state, NO_EVENTS])
+    if (!isKnownStreamBlockType(event.content_block.type)) return Effect.succeed<StepResult>([state, NO_EVENTS])
     const decoded = decodeAnthropicStreamBlock(event.content_block)
     if (Option.isNone(decoded)) return invalidStreamEvent(event)
     const block = decoded.value
@@ -1591,7 +1449,7 @@ const step = (state: ParserState, event: AnthropicEvent) => {
   }
   if (event.type === "content_block_delta") {
     if (!ProviderShared.isRecord(event.delta)) return invalidStreamEvent(event)
-    if (typeof event.delta.type === "string" && !STREAM_DELTA_TYPES.has(event.delta.type))
+    if (typeof event.delta.type === "string" && !isKnownStreamDeltaType(event.delta.type))
       return Effect.succeed<StepResult>([state, NO_EVENTS])
     const decoded = decodeAnthropicStreamDelta(event.delta)
     if (Option.isNone(decoded)) return invalidStreamEvent(event)
@@ -1625,8 +1483,6 @@ export const protocol = Protocol.make({
   stream: {
     event: Protocol.jsonEvent(AnthropicEvent),
     initial: (request) => ({
-      provider: request.model.provider,
-      compactions: {},
       providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
       tools: ToolStream.empty<number>(),
       reasoningSignatures: {},
@@ -1636,37 +1492,6 @@ export const protocol = Protocol.make({
   },
 })
 
-export const transport = <Body extends Pick<AnthropicMessagesBody, "messages" | "context_management">>() => {
-  const http = HttpTransport.httpJson<Body, string>({ framing })
-  return {
-    ...http,
-    prepare: (input: Parameters<typeof http.prepare>[0]) => {
-      if (
-        !input.body.context_management?.edits.length &&
-        !input.body.messages.some((message) => message.content.some((block) => block.type === "compaction"))
-      )
-        return http.prepare(input)
-      const headers = Headers.fromInput(input.request.http?.headers)
-      const betas = new Set(
-        (headers["anthropic-beta"] ?? "")
-          .split(",")
-          .map((item) => item.trim())
-          .filter(Boolean),
-      )
-      betas.add("compact-2026-01-12")
-      return http.prepare({
-        ...input,
-        request: LLMRequest.update(input.request, {
-          http: new HttpOptions({
-            ...input.request.http,
-            headers: { ...headers, "anthropic-beta": [...betas].join(",") },
-          }),
-        }),
-      })
-    },
-  }
-}
-
 export const route = Route.make({
   id: ADAPTER,
   provider: "anthropic",
@@ -1676,7 +1501,7 @@ export const route = Route.make({
     baseURL: DEFAULT_BASE_URL,
   }),
   auth: Auth.none,
-  transport: transport<AnthropicMessagesBody>(),
+  framing,
   headers: () => ({ "anthropic-version": "2023-06-01" }),
 })
 

packages/ai/src/protocols/bedrock-converse.ts

@@ -1,4 +1,4 @@
-import { Effect, Encoding, Schema } from "effect"
+import { Effect, Schema } from "effect"
 import { Route } from "../route/client.js"
 import { Endpoint } from "../route/endpoint.js"
 import { Protocol } from "../route/protocol.js"
@@ -262,28 +262,24 @@ const providerMetadata = (key: string, metadata: Record<string, unknown>): Provi
 
 const reasoningSignature = (part: ReasoningPart, providerMetadataKey: string) => {
   const metadata = part.providerMetadata?.[providerMetadataKey]
-  if (part.encrypted !== undefined) return part.encrypted
-  if (ProviderShared.isRecord(metadata) && typeof metadata.signature === "string") return metadata.signature
+  return (
+    part.encrypted ??
+    (ProviderShared.isRecord(metadata) && typeof metadata.signature === "string" ? metadata.signature : undefined)
+  )
 }
 
 const reasoningRedactedData = (part: ReasoningPart, providerMetadataKey: string) => {
   const metadata = part.providerMetadata?.[providerMetadataKey]
-  if (ProviderShared.isRecord(metadata) && typeof metadata.redactedData === "string") return metadata.redactedData
-}
-
-const removeEmptyToolInputKeys = (input: unknown): unknown => {
-  if (Array.isArray(input)) return input.map(removeEmptyToolInputKeys)
-  if (!ProviderShared.isRecord(input)) return input
-  return Object.fromEntries(
-    Object.entries(input).flatMap(([key, value]) => (key === "" ? [] : [[key, removeEmptyToolInputKeys(value)]])),
-  )
+  return ProviderShared.isRecord(metadata) && typeof metadata.redactedData === "string"
+    ? metadata.redactedData
+    : undefined
 }
 
 const lowerToolCall = (part: ToolCallPart): BedrockToolUseBlock => ({
   toolUse: {
     toolUseId: part.id,
     name: part.name,
-    input: removeEmptyToolInputKeys(part.input),
+    input: part.input,
   },
 })
 
@@ -378,12 +374,6 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
             content.push({ reasoningContent: { redactedContent: redactedData } })
             continue
           }
-          if (signature === undefined || signature.trim().length === 0) {
-            // Interrupted streams and model switches can leave unsigned reasoning.
-            // Preserve readable history as text rather than replay invalid reasoningContent.
-            if (part.text.trim().length > 0) content.push(...textWithCache(breakpoints, part.text, part.cache))
-            continue
-          }
           content.push({ reasoningContent: { reasoningText: { text: part.text, signature } } })
           continue
         }
@@ -392,7 +382,7 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
           continue
         }
       }
-      if (content.length > 0) messages.push({ role: "assistant", content })
+      messages.push({ role: "assistant", content })
       continue
     }
 
@@ -415,56 +405,49 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
 
 // System prompts share the cache-point convention: emit the text block, then
 // optionally a positional `cachePoint` marker.
-const lowerSystem = (breakpoints: BedrockCache.Breakpoints, system: ReadonlyArray<LLMRequest["system"][number]>) => {
-  const content = system
-    .filter((part) => part.text.length > 0)
-    .flatMap((part) => textWithCache(breakpoints, part.text, part.cache))
-  return content.length === 0 ? undefined : content
-}
+const lowerSystem = (
+  breakpoints: BedrockCache.Breakpoints,
+  system: ReadonlyArray<LLMRequest["system"][number]>,
+): BedrockSystemBlock[] => system.flatMap((part) => textWithCache(breakpoints, part.text, part.cache))
 
 const fromRequest = Effect.fn("BedrockConverse.fromRequest")(function* (request: LLMRequest) {
   const toolChoice = request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined
-  const flattened = ProviderShared.flattenToolRequest(request)
   const generation = request.generation
   // Bedrock-Claude shares Anthropic's 4-breakpoint cap. Spend the budget in
   // tools → system → messages order to favour the highest-impact prefixes.
-  const breakpoints = BedrockCache.breakpoints(request.model.id)
-  const toolConfig = (() => {
-    if (flattened.tools.length === 0) return undefined
-    return {
-      tools: lowerTools(request.model.compatibility?.toolSchema, breakpoints, flattened.tools),
-      // Converse has no native "none". Keep definitions stable for prompt
-      // caching and omit only the unsupported choice.
-      toolChoice,
-    }
-  })()
-  const system = lowerSystem(breakpoints, request.system)
-  const messages = yield* lowerMessages(flattened.request, breakpoints)
+  const breakpoints = BedrockCache.breakpoints()
+  const toolConfig =
+    request.tools.length > 0
+      ? {
+          tools: lowerTools(request.model.compatibility?.toolSchema, breakpoints, request.tools),
+          // Converse has no native "none". Keep definitions stable for prompt
+          // caching and omit only the unsupported choice.
+          toolChoice,
+        }
+      : undefined
+  const system = request.system.length === 0 ? undefined : lowerSystem(breakpoints, request.system)
+  const messages = yield* lowerMessages(request, breakpoints)
   if (breakpoints.dropped > 0) {
     yield* Effect.logWarning(
       `Bedrock Converse: dropped ${breakpoints.dropped} cache breakpoint(s); the API allows at most ${BedrockCache.BEDROCK_BREAKPOINT_CAP} per request.`,
     )
   }
-  const inferenceConfig = (() => {
-    if (
-      generation?.maxTokens === undefined &&
-      generation?.temperature === undefined &&
-      generation?.topP === undefined &&
-      (generation?.stop === undefined || generation.stop.length === 0)
-    )
-      return undefined
-    return {
-      maxTokens: generation?.maxTokens,
-      temperature: generation?.temperature,
-      topP: generation?.topP,
-      stopSequences: generation?.stop,
-    }
-  })()
   return {
     modelId: request.model.id,
     messages,
     system,
-    inferenceConfig,
+    inferenceConfig:
+      generation?.maxTokens === undefined &&
+      generation?.temperature === undefined &&
+      generation?.topP === undefined &&
+      (generation?.stop === undefined || generation.stop.length === 0)
+        ? undefined
+        : {
+            maxTokens: generation?.maxTokens,
+            temperature: generation?.temperature,
+            topP: generation?.topP,
+            stopSequences: generation?.stop,
+          },
     toolConfig,
     // Converse's base inferenceConfig has no topK; Anthropic/Nova accept it
     // as a model-specific field, so it goes through additionalModelRequestFields.
@@ -480,6 +463,7 @@ const mapFinishReason = (reason: string): FinishReason => {
   if (reason === "max_tokens" || reason === "model_context_window_exceeded") return "length"
   if (reason === "tool_use") return "tool-calls"
   if (reason === "content_filtered" || reason === "guardrail_intervened") return "content-filter"
+  if (reason === "malformed_model_output" || reason === "malformed_tool_use") return "error"
   return "unknown"
 }
 
@@ -507,23 +491,12 @@ interface ParserState {
   readonly providerMetadataKey: string
   readonly tools: ToolStream.State<number>
   // Bedrock splits the finish into `messageStop` (carries `stopReason`) and
-  // `metadata` (carries usage). Hold both in state so `onHalt` can emit exactly
-  // one finish after both chunks have had a chance to arrive.
-  readonly finishReason: FinishReasonDetails | undefined
-  readonly usage: Usage | undefined
+  // `metadata` (carries usage). Hold the terminal event in state so `onHalt`
+  // can emit exactly one finish after both chunks have had a chance to arrive.
+  readonly pendingFinish: { readonly reason: FinishReasonDetails; readonly usage?: Usage } | undefined
   readonly hasToolCalls: boolean
   readonly lifecycle: Lifecycle.State
   readonly reasoningSignatures: Readonly<Record<number, string>>
-  readonly reasoningRedactedContent: Readonly<Record<number, ReadonlyArray<Uint8Array>>>
-}
-
-const encodeRedactedContent = (chunks: ReadonlyArray<Uint8Array>) => {
-  const bytes = new Uint8Array(chunks.reduce((total, chunk) => total + chunk.length, 0))
-  chunks.reduce((offset, chunk) => {
-    bytes.set(chunk, offset)
-    return offset + chunk.length
-  }, 0)
-  return Encoding.encodeBase64(bytes)
 }
 
 const step = (state: ParserState, event: BedrockEvent) =>
@@ -571,60 +544,38 @@ const step = (state: ParserState, event: BedrockEvent) =>
       const index = event.contentBlockDelta.contentBlockIndex
       const reasoning = event.contentBlockDelta.delta.reasoningContent
       const events: LLMEvent[] = []
-      const redactedChunks = yield* (() => {
-        if (reasoning.redactedContent === undefined) return Effect.succeed(undefined)
-        return Effect.fromResult(Encoding.decodeBase64(reasoning.redactedContent)).pipe(
-          Effect.map((chunk) => [...(state.reasoningRedactedContent[index] ?? []), chunk]),
-          Effect.mapError((cause) =>
-            ProviderShared.eventError(
-              ADAPTER,
-              "Bedrock Converse reasoningContent.redactedContent contains invalid base64 data",
-              undefined,
-              cause,
-            ),
-          ),
-        )
-      })()
-      const redactedData = redactedChunks === undefined ? reasoning.data : encodeRedactedContent(redactedChunks)
-      const metadata = (() => {
-        if (reasoning.signature) return providerMetadata(state.providerMetadataKey, { signature: reasoning.signature })
-        if (redactedData !== undefined) return providerMetadata(state.providerMetadataKey, { redactedData })
-      })()
-      const lifecycle = (() => {
-        if (reasoning.text === undefined && metadata === undefined) return state.lifecycle
-        return Lifecycle.reasoningDelta(state.lifecycle, events, `reasoning-${index}`, reasoning.text ?? "", metadata)
-      })()
-      const reasoningRedactedContent = (() => {
-        if (redactedChunks !== undefined) return { ...state.reasoningRedactedContent, [index]: redactedChunks }
-        if (reasoning.data === undefined) return state.reasoningRedactedContent
-        return Object.fromEntries(
-          Object.entries(state.reasoningRedactedContent).filter(([key]) => key !== String(index)),
-        )
-      })()
-      const reasoningSignatures = (() => {
-        if (!reasoning.signature) return state.reasoningSignatures
-        return { ...state.reasoningSignatures, [index]: reasoning.signature }
-      })()
+      const redactedData = reasoning.redactedContent ?? reasoning.data
+      const metadata = reasoning.signature
+        ? providerMetadata(state.providerMetadataKey, { signature: reasoning.signature })
+        : redactedData !== undefined
+          ? providerMetadata(state.providerMetadataKey, { redactedData })
+          : undefined
+      const lifecycle =
+        reasoning.text !== undefined || metadata !== undefined
+          ? Lifecycle.reasoningDelta(state.lifecycle, events, `reasoning-${index}`, reasoning.text ?? "", metadata)
+          : state.lifecycle
       return [
         {
           ...state,
           lifecycle,
-          reasoningSignatures,
-          reasoningRedactedContent,
+          reasoningSignatures: reasoning.signature
+            ? { ...state.reasoningSignatures, [index]: reasoning.signature }
+            : state.reasoningSignatures,
         },
         events,
       ] as const
     }
 
     if (event.contentBlockDelta?.delta?.toolUse) {
-      // A delta for a block that is not open, whether it already stopped or never
-      // started, has nothing to attach to and is dropped.
-      const result = ToolStream.append(
+      const index = event.contentBlockDelta.contentBlockIndex
+      const result = ToolStream.appendExisting(
+        ADAPTER,
         state.tools,
-        event.contentBlockDelta.contentBlockIndex,
+        index,
         event.contentBlockDelta.delta.toolUse.input,
+        "Bedrock Converse tool delta is missing its tool call",
       )
-      if (!result) return [state, []] as const
+      if (ToolStream.isError(result)) return yield* result
       const events: LLMEvent[] = []
       const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
       events.push(...result.events)
@@ -636,24 +587,16 @@ const step = (state: ParserState, event: BedrockEvent) =>
       const result = yield* ToolStream.finish(ADAPTER, state.tools, index)
       const events: LLMEvent[] = []
       const resultEvents = result.events ?? []
-      const lifecycle = (() => {
-        if (resultEvents.length) return Lifecycle.stepStart(state.lifecycle, events)
-        const metadata = (() => {
-          const signature = state.reasoningSignatures[index]
-          if (signature) return providerMetadata(state.providerMetadataKey, { signature })
-          const redactedContent = state.reasoningRedactedContent[index]
-          if (redactedContent)
-            return providerMetadata(state.providerMetadataKey, {
-              redactedData: encodeRedactedContent(redactedContent),
-            })
-        })()
-        return Lifecycle.reasoningEnd(
-          Lifecycle.textEnd(state.lifecycle, events, `text-${index}`),
-          events,
-          `reasoning-${index}`,
-          metadata,
-        )
-      })()
+      const lifecycle = resultEvents.length
+        ? Lifecycle.stepStart(state.lifecycle, events)
+        : Lifecycle.reasoningEnd(
+            Lifecycle.textEnd(state.lifecycle, events, `text-${index}`),
+            events,
+            `reasoning-${index}`,
+            state.reasoningSignatures[index]
+              ? providerMetadata(state.providerMetadataKey, { signature: state.reasoningSignatures[index] })
+              : undefined,
+          )
       events.push(...resultEvents)
       return [
         {
@@ -666,30 +609,21 @@ const step = (state: ParserState, event: BedrockEvent) =>
           reasoningSignatures: Object.fromEntries(
             Object.entries(state.reasoningSignatures).filter(([key]) => key !== String(index)),
           ),
-          reasoningRedactedContent: Object.fromEntries(
-            Object.entries(state.reasoningRedactedContent).filter(([key]) => key !== String(index)),
-          ),
         },
         events,
       ] as const
     }
 
     if (event.messageStop) {
-      if (
-        event.messageStop.stopReason === "malformed_model_output" ||
-        event.messageStop.stopReason === "malformed_tool_use"
-      )
-        return yield* ProviderShared.eventError(
-          ADAPTER,
-          `Bedrock Converse stopped with ${event.messageStop.stopReason}`,
-          ProviderShared.encodeJson(event),
-        )
       return [
         {
           ...state,
-          finishReason: {
-            normalized: mapFinishReason(event.messageStop.stopReason),
-            raw: event.messageStop.stopReason,
+          pendingFinish: {
+            reason: {
+              normalized: mapFinishReason(event.messageStop.stopReason),
+              raw: event.messageStop.stopReason,
+            },
+            usage: state.pendingFinish?.usage,
           },
         },
         [],
@@ -697,11 +631,14 @@ const step = (state: ParserState, event: BedrockEvent) =>
     }
 
     if (event.metadata) {
-      const usage = mapUsage(event.metadata.usage, state.providerMetadataKey) ?? state.usage
+      const usage = mapUsage(event.metadata.usage, state.providerMetadataKey) ?? state.pendingFinish?.usage
       return [
         {
           ...state,
-          usage,
+          pendingFinish: {
+            reason: state.pendingFinish?.reason ?? { normalized: "stop" },
+            usage,
+          },
         },
         [],
       ] as const
@@ -724,22 +661,23 @@ const step = (state: ParserState, event: BedrockEvent) =>
 
 const framing = BedrockEventStream.framing(ADAPTER)
 
-const onHalt = (state: ParserState): ReadonlyArray<LLMEvent> => {
-  if (!state.finishReason) return []
-  const normalized = (() => {
-    if (state.finishReason.normalized === "stop" && state.hasToolCalls) return "tool-calls"
-    return state.finishReason.normalized
-  })()
-  const events: LLMEvent[] = []
-  Lifecycle.finish(state.lifecycle, events, {
-    reason: {
-      ...state.finishReason,
-      normalized,
-    },
-    usage: state.usage,
-  })
-  return events
-}
+const onHalt = (state: ParserState): ReadonlyArray<LLMEvent> =>
+  state.pendingFinish
+    ? (() => {
+        const events: LLMEvent[] = []
+        Lifecycle.finish(state.lifecycle, events, {
+          reason: {
+            ...state.pendingFinish.reason,
+            normalized:
+              state.pendingFinish.reason.normalized === "stop" && state.hasToolCalls
+                ? "tool-calls"
+                : state.pendingFinish.reason.normalized,
+          },
+          usage: state.pendingFinish.usage,
+        })
+        return events
+      })()
+    : []
 
 // =============================================================================
 // Protocol And Bedrock Route
@@ -759,12 +697,10 @@ export const protocol = Protocol.make({
     initial: (request) => ({
       providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
       tools: ToolStream.empty<number>(),
-      finishReason: undefined,
-      usage: undefined,
+      pendingFinish: undefined,
       hasToolCalls: false,
       lifecycle: Lifecycle.initial(),
       reasoningSignatures: {},
-      reasoningRedactedContent: {},
     }),
     step,
     onHalt: (state) => Effect.succeed(onHalt(state)),

packages/ai/src/protocols/bedrock-event-stream.ts

@@ -1,7 +1,7 @@
 import { EventStreamCodec } from "@smithy/eventstream-codec"
 import { fromUtf8, toUtf8 } from "@smithy/util-utf8"
 import { Effect, Encoding, Stream } from "effect"
-import { AIError, AIErrorReason, InvalidProviderOutputError } from "../schema/index.js"
+import { AIError, AIErrorReason } from "../schema/index.js"
 import { Framing } from "../route/framing.js"
 import { ProviderShared } from "./shared.js"
 
@@ -22,10 +22,6 @@ interface FrameBufferState {
 
 const initialFrameBuffer: FrameBufferState = { buffer: new Uint8Array(0), offset: 0 }
 
-type FrameInput = { readonly _tag: "Chunk"; readonly bytes: Uint8Array } | { readonly _tag: "End" }
-
-const endOfStream: FrameInput = { _tag: "End" }
-
 const appendChunk = (state: FrameBufferState, chunk: Uint8Array): FrameBufferState => {
   const remaining = state.buffer.length - state.offset
   // Compact: drop the consumed prefix and append the new chunk in one alloc.
@@ -37,23 +33,9 @@ const appendChunk = (state: FrameBufferState, chunk: Uint8Array): FrameBufferSta
   return { buffer: next, offset: 0 }
 }
 
-const consumeFrames = (route: string) => (state: FrameBufferState, input: FrameInput) =>
+const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8Array) =>
   Effect.gen(function* () {
-    if (input._tag === "End") {
-      const remaining = state.buffer.subarray(state.offset)
-      if (remaining.length > 0)
-        return yield* new AIError({
-          reason: new InvalidProviderOutputError({
-            route,
-            classification: "incomplete-stream",
-            message: `Incomplete Bedrock Converse event-stream frame: ${remaining.length} buffered bytes remain at end of stream`,
-            body: Encoding.encodeBase64(remaining),
-          }),
-        })
-      return [state, []] as const
-    }
-
-    let cursor = appendChunk(state, input.bytes)
+    let cursor = appendChunk(state, chunk)
     const out: object[] = []
     while (cursor.buffer.length - cursor.offset >= 4) {
       const view = cursor.buffer.subarray(cursor.offset)
@@ -131,12 +113,7 @@ const consumeFrames = (route: string) => (state: FrameBufferState, input: FrameI
 export const framing = (route: string): Framing.Definition<object> => ({
   id: "aws-event-stream",
   body: (frame) => ("rawBody" in frame && typeof frame.rawBody === "string" ? frame.rawBody : undefined),
-  frame: (bytes) =>
-    bytes.pipe(
-      Stream.map((bytes): FrameInput => ({ _tag: "Chunk", bytes })),
-      Stream.concat(Stream.succeed(endOfStream)),
-      Stream.mapAccumEffect(() => initialFrameBuffer, consumeFrames(route)),
-    ),
+  frame: (bytes) => bytes.pipe(Stream.mapAccumEffect(() => initialFrameBuffer, consumeFrames(route))),
 })
 
 export * as BedrockEventStream from "./bedrock-event-stream.js"

packages/ai/src/protocols/gemini.ts

@@ -1,5 +1,5 @@
 import { Effect, Option, Schema } from "effect"
-import { Tool } from "@opencode/schema/tool"
+import { Tool } from "@opencode-ai/schema/tool"
 import { Route } from "../route/client.js"
 import { Auth } from "../route/auth.js"
 import { Endpoint } from "../route/endpoint.js"
@@ -240,10 +240,6 @@ interface ParserState {
   readonly lifecycle: Lifecycle.State
   readonly reasoningSignature?: string
   readonly textSignature?: string
-  readonly reasoningId?: string
-  readonly textId?: string
-  readonly nextReasoningId: number
-  readonly nextTextId: number
   readonly seenCallIds?: ReadonlySet<string>
 }
 
@@ -465,8 +461,7 @@ function mapSafetySettings(value: unknown) {
 }
 
 const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMRequest) {
-  const flattened = ProviderShared.flattenToolRequest(request)
-  const hasTools = flattened.tools.length > 0
+  const hasTools = request.tools.length > 0
   const generation = request.generation
   const options = resolveOptions(request)
   const toolSchemaCompatibility = request.model.compatibility?.toolSchema
@@ -484,7 +479,7 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
 
   return {
     cachedContent: options.cachedContent,
-    contents: yield* lowerMessages(flattened.request),
+    contents: yield* lowerMessages(request),
     safetySettings: options.safetySettings,
     serviceTier: options.serviceTier,
     systemInstruction:
@@ -492,7 +487,7 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
     tools: hasTools
       ? [
           {
-            functionDeclarations: flattened.tools.map((tool) =>
+            functionDeclarations: request.tools.map((tool) =>
               lowerTool(tool, ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility)),
             ),
           },
@@ -576,23 +571,19 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
 
   const events: LLMEvent[] = []
   let lifecycle = state.lifecycle
-  if (state.reasoningId !== undefined)
+  if (state.reasoningSignature !== undefined)
     lifecycle = Lifecycle.reasoningEnd(
       lifecycle,
       events,
-      state.reasoningId,
-      state.reasoningSignature === undefined
-        ? undefined
-        : providerMetadata(state.providerMetadataKey, { thoughtSignature: state.reasoningSignature }),
+      "reasoning-0",
+      providerMetadata(state.providerMetadataKey, { thoughtSignature: state.reasoningSignature }),
     )
-  if (state.textId !== undefined)
+  if (state.textSignature !== undefined)
     lifecycle = Lifecycle.textEnd(
       lifecycle,
       events,
-      state.textId,
-      state.textSignature === undefined
-        ? undefined
-        : providerMetadata(state.providerMetadataKey, { thoughtSignature: state.textSignature }),
+      "text-0",
+      providerMetadata(state.providerMetadataKey, { thoughtSignature: state.textSignature }),
     )
   Lifecycle.finish(lifecycle, events, {
     reason: {
@@ -610,27 +601,18 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
 }
 
 const step = (state: ParserState, event: GeminiEvent) => {
-  if (ProviderShared.isRecord(event.error)) {
+  if (ProviderShared.isRecord(event.error) && typeof event.error.message === "string") {
     const body = ProviderShared.encodeJson(event)
     return Effect.fail(
       new AIError({
         reason: classifyProviderFailure({
-          message:
-            typeof event.error.message === "string" && event.error.message.length > 0
-              ? event.error.message
-              : typeof event.error.status === "string" && event.error.status.length > 0
-                ? event.error.status
-                : "Gemini provider error",
+          message: event.error.message,
           status: typeof event.error.code === "number" ? event.error.code : undefined,
           rawBody: body,
         }),
       }),
     )
   }
-  if ("error" in event)
-    return Effect.fail(
-      ProviderShared.eventError(state.route, `Invalid ${state.route} stream event`, ProviderShared.encodeJson(event)),
-    )
   const nextState = {
     ...state,
     promptFeedback: event.promptFeedback ?? state.promptFeedback,
@@ -639,14 +621,6 @@ const step = (state: ParserState, event: GeminiEvent) => {
       : state.usage,
   }
   const candidate = event.candidates?.[0]
-  if (candidate?.finishReason && mapFinishReason(candidate.finishReason, state.hasToolCalls) === "error")
-    return Effect.fail(
-      ProviderShared.eventError(
-        state.route,
-        `Gemini stopped with ${candidate.finishReason}`,
-        ProviderShared.encodeJson(event),
-      ),
-    )
   if (!candidate?.content)
     return Effect.succeed([
       { ...nextState, finishReason: candidate?.finishReason ?? nextState.finishReason },
@@ -658,10 +632,6 @@ const step = (state: ParserState, event: GeminiEvent) => {
   let lifecycle = nextState.lifecycle
   let reasoningSignature = nextState.reasoningSignature
   let textSignature = nextState.textSignature
-  let reasoningId = nextState.reasoningId
-  let textId = nextState.textId
-  let nextReasoningId = nextState.nextReasoningId
-  let nextTextId = nextState.nextTextId
   // Supplier ids must be tracked across chunks of the same response, not just within one event's parts.
   const seenCallIds = new Set(nextState.seenCallIds)
 
@@ -687,51 +657,27 @@ const step = (state: ParserState, event: GeminiEvent) => {
     else if (signature !== undefined && "text" in part) textSignature = signature
     if ("text" in part && part.text.length > 0) {
       if (part.thought) {
-        if (textId !== undefined) {
-          lifecycle = Lifecycle.textEnd(
-            lifecycle,
-            events,
-            textId,
-            textSignature
-              ? providerMetadata(state.providerMetadataKey, { thoughtSignature: textSignature })
-              : undefined,
-          )
-          textId = undefined
-          textSignature = undefined
-        }
-        if (reasoningId === undefined) {
-          reasoningId = `reasoning-${nextReasoningId}`
-          nextReasoningId += 1
-        }
         lifecycle = Lifecycle.reasoningDelta(
           lifecycle,
           events,
-          reasoningId,
+          "reasoning-0",
           part.text,
           signature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: signature }) : undefined,
         )
         continue
       }
-      if (reasoningId !== undefined) {
-        lifecycle = Lifecycle.reasoningEnd(
-          lifecycle,
-          events,
-          reasoningId,
-          reasoningSignature
-            ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-            : undefined,
-        )
-        reasoningId = undefined
-        reasoningSignature = undefined
-      }
-      if (textId === undefined) {
-        textId = `text-${nextTextId}`
-        nextTextId += 1
-      }
+      lifecycle = Lifecycle.reasoningEnd(
+        lifecycle,
+        events,
+        "reasoning-0",
+        reasoningSignature
+          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
+          : undefined,
+      )
       lifecycle = Lifecycle.textDelta(
         lifecycle,
         events,
-        textId,
+        "text-0",
         part.text,
         textSignature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: textSignature }) : undefined,
       )
@@ -749,28 +695,14 @@ const step = (state: ParserState, event: GeminiEvent) => {
       const duplicate = supplied !== undefined && seenCallIds.has(supplied)
       if (supplied !== undefined) seenCallIds.add(supplied)
       const id = supplied !== undefined && !duplicate ? supplied : `tool_${crypto.randomUUID().replaceAll("-", "")}`
-      if (reasoningId !== undefined) {
-        lifecycle = Lifecycle.reasoningEnd(
-          lifecycle,
-          events,
-          reasoningId,
-          reasoningSignature
-            ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-            : undefined,
-        )
-        reasoningId = undefined
-        reasoningSignature = undefined
-      }
-      if (textId !== undefined) {
-        lifecycle = Lifecycle.textEnd(
-          lifecycle,
-          events,
-          textId,
-          textSignature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: textSignature }) : undefined,
-        )
-        textId = undefined
-        textSignature = undefined
-      }
+      lifecycle = Lifecycle.reasoningEnd(
+        lifecycle,
+        events,
+        "reasoning-0",
+        reasoningSignature
+          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
+          : undefined,
+      )
       lifecycle = Lifecycle.stepStart(lifecycle, events)
       events.push(
         LLMEvent.toolCall({
@@ -793,10 +725,6 @@ const step = (state: ParserState, event: GeminiEvent) => {
       lifecycle,
       reasoningSignature,
       textSignature,
-      reasoningId,
-      textId,
-      nextReasoningId,
-      nextTextId,
       seenCallIds,
       finishReason: candidate.finishReason ?? nextState.finishReason,
     },
@@ -824,8 +752,6 @@ export const protocol = Protocol.make({
       providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
       hasToolCalls: false,
       lifecycle: Lifecycle.initial(),
-      nextReasoningId: 0,
-      nextTextId: 0,
     }),
     step,
     onHalt: (state) => Effect.succeed(finish(state)),

packages/ai/src/protocols/index.ts

@@ -1,7 +1,6 @@
 export * as AnthropicMessages from "./anthropic-messages.js"
 export * as BedrockConverse from "./bedrock-converse.js"
 export * as Gemini from "./gemini.js"
-export * as MistralChat from "./mistral-chat.js"
 export * as OpenAIChat from "./openai-chat.js"
 export * as OpenAIImages from "./openai-images.js"
 export * as OpenAICompatibleChat from "./openai-compatible-chat.js"

packages/ai/src/protocols/meta-images.ts

@@ -1,133 +0,0 @@
-import { Effect, Encoding, Schema } from "effect"
-import { Headers, HttpClientRequest } from "effect/unstable/http"
-import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image.js"
-import { Auth } from "../route/auth.js"
-import { Usage, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema/index.js"
-import { JsonObject, ProviderShared, optionalNull } from "./shared.js"
-import { ImageInputs } from "./utils/image-input.js"
-
-type OpenString<Known extends string> = Known | (string & {})
-export type ImageOptions = {
-  readonly n?: number
-  /** Aspect ratio hint, not an exact output resolution. */
-  readonly size?: string
-  readonly outputFormat?: OpenString<"webp" | "png" | "jpeg">
-  readonly responseFormat?: OpenString<"b64_json" | "url">
-  readonly reasoningStrength?: OpenString<"low" | "high">
-  readonly toolEnablement?: {
-    readonly enable_image_search?: boolean
-    readonly enable_web_search?: boolean
-    readonly enable_shell?: boolean
-  }
-  readonly [key: string]: unknown
-}
-
-const Body = Schema.StructWithRest(
-  Schema.Struct({
-    model: Schema.String,
-    prompt: Schema.String,
-    images: Schema.optional(Schema.Array(JsonObject)),
-    n: Schema.optional(Schema.Number),
-    size: Schema.optional(Schema.String),
-    output_format: Schema.optional(Schema.String),
-    response_format: Schema.optional(Schema.String),
-    reasoning_strength: Schema.optional(Schema.String),
-    tool_enablement: Schema.optional(Schema.Record(Schema.String, Schema.Boolean)),
-  }),
-  [JsonObject],
-)
-
-const Response = Schema.Struct({
-  data: Schema.Array(Schema.Struct({ b64_json: optionalNull(Schema.String), url: optionalNull(Schema.String) })),
-  output_format: Schema.optional(Schema.String),
-  usage: Schema.optional(
-    Schema.Struct({
-      input_tokens: Schema.optional(Schema.Number),
-      output_tokens: Schema.optional(Schema.Number),
-      total_tokens: Schema.optional(Schema.Number),
-    }),
-  ),
-})
-
-export const model = (input: {
-  readonly id: string
-  readonly auth: Auth.Definition
-  readonly baseURL: string
-  readonly headers?: Record<string, string>
-  readonly http?: HttpOptions
-}) => {
-  const route: ImageRoute<ImageOptions> = {
-    id: "meta-images",
-    generate: Effect.fn("MetaImages.generate")(function* (request: ImageRequestFor<ImageOptions>, execute) {
-      const http = mergeHttpOptions(request.model.http, request.http)
-      const images = yield* Effect.forEach(request.images ?? [], (image) => {
-        if (image.type === "bytes") return Effect.succeed({ image_url: ImageInputs.dataUrl(image) })
-        if (image.type === "url") return Effect.succeed({ image_url: image.url })
-        return ImageInputs.invalid("Meta Images accepts image bytes and URLs")
-      })
-      const { outputFormat, responseFormat, reasoningStrength, toolEnablement, ...native } = request.options ?? {}
-      const payload = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(Body))(
-        mergeJsonRecords(
-          {
-            model: request.model.id,
-            prompt: request.prompt,
-            images: images.length === 0 ? undefined : images,
-            output_format: outputFormat,
-            response_format: responseFormat,
-            reasoning_strength: reasoningStrength,
-            tool_enablement: toolEnablement,
-          },
-          native,
-          http?.body,
-        ),
-      )
-      const body = ProviderShared.encodeJson(payload)
-      const url = new URL(`${input.baseURL.replace(/\/$/, "")}/images/${images.length === 0 ? "generations" : "edits"}`)
-      Object.entries(http?.query ?? {}).forEach(([key, value]) => url.searchParams.set(key, value))
-      const headers = yield* Auth.toEffect(input.auth)({
-        request,
-        method: "POST",
-        url: url.toString(),
-        body,
-        headers: Headers.fromInput({ ...input.headers, ...http?.headers }),
-      })
-      const response = yield* execute(
-        HttpClientRequest.post(url.toString()).pipe(
-          HttpClientRequest.setHeaders(headers),
-          HttpClientRequest.bodyText(body, "application/json"),
-        ),
-      )
-      const output = yield* ProviderShared.imageResponse("meta-images", "Meta Images", response)
-      const decoded = yield* Schema.decodeUnknownEffect(Schema.fromJsonString(Response))(output.body).pipe(
-        Effect.mapError((cause) => output.invalid("Meta Images returned an invalid response", cause)),
-      )
-      const format = decoded.output_format ?? payload.output_format ?? "webp"
-      const generated = yield* Effect.forEach(decoded.data, (item, index) => {
-        if (item.b64_json)
-          return Effect.fromResult(Encoding.decodeBase64(item.b64_json)).pipe(
-            Effect.mapError((cause) => output.invalid(`Meta Images result ${index} contains invalid base64`, cause)),
-            Effect.map((data) => new GeneratedImage({ mediaType: `image/${format}`, data })),
-          )
-        if (item.url) return Effect.succeed(new GeneratedImage({ mediaType: `image/${format}`, data: item.url }))
-        return output.invalid(`Meta Images result ${index} has neither image data nor a URL`)
-      })
-      if (generated.length === 0) return yield* output.invalid("Meta Images returned no images")
-      return new ImageResponse({
-        images: generated,
-        usage:
-          decoded.usage === undefined
-            ? undefined
-            : new Usage({
-                inputTokens: decoded.usage.input_tokens,
-                outputTokens: decoded.usage.output_tokens,
-                totalTokens: decoded.usage.total_tokens,
-                providerMetadata: { meta: decoded.usage },
-              }),
-        providerMetadata: { meta: { outputFormat: format } },
-      })
-    }),
-  }
-  return ImageModel.make<ImageOptions>({ id: input.id, provider: "meta", route, http: input.http })
-}
-
-export * as MetaImages from "./meta-images.js"

packages/ai/src/protocols/meta-messages.ts

@@ -1,52 +0,0 @@
-import { Effect, Schema } from "effect"
-import { Protocol } from "../route/protocol.js"
-import type { LLMRequest } from "../schema/index.js"
-import { AnthropicMessages } from "./anthropic-messages.js"
-import { MetaResponses } from "./meta-responses.js"
-import { JsonObject, optionalArray, ProviderShared } from "./shared.js"
-
-const WebSearch = Schema.Struct({
-  type: Schema.Literal("web_search"),
-  name: Schema.Literal("web_search"),
-  user_location: MetaResponses.WebSearch.fields.user_location,
-})
-const Body = Schema.Struct({
-  ...AnthropicMessages.AnthropicMessagesBody.fields,
-  tools: optionalArray(
-    Schema.Union([
-      Schema.Struct({ name: Schema.String, description: Schema.String, input_schema: JsonObject }),
-      WebSearch,
-    ]),
-  ),
-})
-
-const fromRequest = Effect.fn("MetaMessages.fromRequest")(function* (request: LLMRequest) {
-  const projected = ProviderShared.flattenToolRequest(request)
-  const body = yield* AnthropicMessages.protocol.body.from(projected.request)
-  return {
-    ...body,
-    tools:
-      body.tools === undefined
-        ? undefined
-        : yield* Effect.forEach(body.tools, (tool, index) =>
-            Effect.gen(function* () {
-              const native = projected.tools[index]?.native
-              if (native === undefined) return tool
-              const search = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(MetaResponses.WebSearch))(
-                native.meta,
-              )
-              if (search.search_context_size !== undefined)
-                return yield* ProviderShared.invalidRequest("Meta Messages does not support searchContextSize")
-              return { type: "web_search" as const, name: "web_search" as const, user_location: search.user_location }
-            }),
-          ),
-  }
-})
-
-export const protocol = Protocol.make({
-  id: "meta-messages",
-  body: { schema: Body, from: fromRequest },
-  stream: AnthropicMessages.protocol.stream,
-})
-
-export * as MetaMessages from "./meta-messages.js"

packages/ai/src/protocols/meta-responses.ts

@@ -1,238 +0,0 @@
-import { Effect, Encoding, Schema } from "effect"
-import { Protocol } from "../route/protocol.js"
-import { HttpTransport } from "../route/transport/index.js"
-import { LLMEvent, LLMRequest, Message, ToolResultPart } from "../schema/index.js"
-import { OpenResponses } from "./open-responses.js"
-import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
-import { ResponsesHostedTools } from "./utils/responses-hosted-tools.js"
-import { ToolSchemaProjection } from "./utils/tool-schema.js"
-import { MetaImage } from "./utils/meta-image.js"
-
-const ADAPTER = "meta-responses"
-const NAME = "Meta Responses"
-
-export const WebSearch = Schema.Struct({
-  type: Schema.Literal("web_search"),
-  search_context_size: Schema.optional(Schema.String),
-  user_location: Schema.optional(
-    Schema.Struct({
-      type: Schema.Literal("approximate"),
-      city: Schema.optional(Schema.String),
-      region: Schema.optional(Schema.String),
-      country: Schema.optional(Schema.String),
-      timezone: Schema.optional(Schema.String),
-    }),
-  ),
-})
-
-export const ImageGeneration = Schema.Struct({
-  type: Schema.Literal("image_generation"),
-  size: Schema.optional(Schema.String),
-  output_format: Schema.optional(Schema.String),
-  reasoning_strength: Schema.optional(Schema.String),
-  enable_image_search: Schema.optional(Schema.Boolean),
-  enable_web_search: Schema.optional(Schema.Boolean),
-  enable_shell: Schema.optional(Schema.Boolean),
-})
-
-const NativeTool = Schema.Union([WebSearch, ImageGeneration])
-const ImageItem = Schema.Struct({
-  type: Schema.Literal("image_generation_call"),
-  id: Schema.String,
-  status: Schema.optional(Schema.String),
-  result: optionalNull(Schema.String),
-  output_format: Schema.optional(Schema.String),
-  error: Schema.optional(Schema.Unknown),
-})
-
-const Body = Schema.Struct({
-  ...OpenResponses.coreFields,
-  input: Schema.Array(Schema.Union([OpenResponses.InputItem, ImageItem])),
-  tools: optionalArray(Schema.Union([OpenResponses.Tool, NativeTool])),
-  stream: Schema.Literal(true),
-})
-
-const MessageAnnotations = Schema.Struct({
-  content: Schema.Array(Schema.Struct({ annotations: optionalArray(JsonObject) })),
-})
-
-interface ParserState extends OpenResponses.ParserState {
-  readonly completedItems: ReadonlySet<string>
-}
-
-const adapter = {
-  id: ADAPTER,
-  name: NAME,
-  restoreHostedToolItem: (item: unknown) => (Schema.is(ImageItem)(item) ? item : undefined),
-} satisfies OpenResponses.ProviderAdapter
-
-const fromRequest = Effect.fn("MetaResponses.fromRequest")(function* (request: LLMRequest) {
-  const key = request.model.route.providerMetadataKey ?? String(request.model.provider)
-  const projected = ProviderShared.flattenToolRequest(
-    LLMRequest.update(request, {
-      messages: request.messages.map((message) =>
-        Message.make({
-          ...message,
-          content: message.content.map((part) => {
-            if (
-              part.type !== "tool-result" ||
-              !part.providerExecuted ||
-              part.name !== "image_generation" ||
-              part.result.type !== "content" ||
-              part.providerMetadata?.[key]?.itemId !== part.id
-            )
-              return part
-            // Meta's signed image ID carries edit state; replay the handle, not the image bytes as a user message.
-            return ToolResultPart.make({
-              ...part,
-              result: {
-                type: "json",
-                value: { type: "image_generation_call", id: part.id, status: "completed", result: null },
-              },
-            })
-          }),
-        }),
-      ),
-    }),
-  )
-  return yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(Body))({
-    ...(yield* OpenResponses.lowerConversation(projected.request, adapter)),
-    ...OpenResponses.lowerGeneration(request),
-    tools:
-      projected.tools.length === 0
-        ? undefined
-        : yield* Effect.forEach(projected.tools, (tool) =>
-            Effect.gen(function* () {
-              if (tool.native === undefined)
-                return yield* OpenResponses.lowerTool(
-                  NAME,
-                  tool,
-                  ToolSchemaProjection.modelCompatibility(tool.inputSchema, request.model.compatibility?.toolSchema),
-                )
-              return yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(NativeTool))(tool.native.meta)
-            }),
-          ),
-    tool_choice:
-      OpenResponses.allowedToolChoice(request) ??
-      (request.toolChoice ? yield* OpenResponses.lowerToolChoice(NAME, request.toolChoice) : undefined),
-  })
-})
-
-const HOSTED_TOOLS = {
-  web_search_call: { name: "web_search", input: (item) => item.action ?? {} },
-  image_generation_call: {
-    name: "image_generation",
-    input: () => ({}),
-    result: Effect.fn("MetaResponses.imageResult")(function* (raw: ResponsesHostedTools.Item) {
-      const item = yield* Schema.decodeUnknownEffect(ImageItem)(raw).pipe(
-        Effect.mapError((cause) =>
-          ProviderShared.eventError(
-            ADAPTER,
-            "Meta returned an invalid image item",
-            ProviderShared.encodeJson(raw),
-            cause,
-          ),
-        ),
-      )
-      if (item.error !== undefined && item.error !== null) return { type: "error" as const, value: item.error }
-      if (!item.result)
-        return yield* ProviderShared.eventError(
-          ADAPTER,
-          "Meta returned an image without data",
-          ProviderShared.encodeJson(raw),
-        )
-      const data = yield* Effect.fromResult(Encoding.decodeBase64(item.result)).pipe(
-        Effect.mapError((cause) =>
-          ProviderShared.eventError(
-            ADAPTER,
-            "Meta returned invalid image base64",
-            ProviderShared.encodeJson(raw),
-            cause,
-          ),
-        ),
-      )
-      const mime = MetaImage.mediaType(data, item.output_format)
-      return {
-        type: "content" as const,
-        value: [{ type: "file" as const, uri: `data:${mime};base64,${item.result}`, mime }],
-      }
-    }),
-  },
-} satisfies ResponsesHostedTools.Definitions
-
-const onEvent = Effect.fn("MetaResponses.onEvent")(function* (
-  state: OpenResponses.ParserState,
-  input: OpenResponses.Event,
-) {
-  const event = OpenResponses.normalize(state, input)
-  if (event.type === "response.output_item.done" && event.item && ResponsesHostedTools.isItem(event.item, HOSTED_TOOLS))
-    return yield* ResponsesHostedTools.onDone(state, event.item, HOSTED_TOOLS)
-  const result = yield* OpenResponses.step(state, event)
-  if (event.type !== "response.output_item.done" || event.item?.type !== "message") return result
-  const message = yield* Schema.decodeUnknownEffect(MessageAnnotations)(event.item).pipe(
-    Effect.mapError((cause) =>
-      ProviderShared.eventError(
-        ADAPTER,
-        "Meta returned invalid message annotations",
-        ProviderShared.encodeJson(event),
-        cause,
-      ),
-    ),
-  )
-  const annotations = message.content.flatMap((part) => part.annotations ?? [])
-  if (annotations.length === 0) return result
-  return [
-    result[0],
-    result[1].map((item) =>
-      LLMEvent.is.textEnd(item)
-        ? LLMEvent.textEnd({
-            ...item,
-            providerMetadata: {
-              ...item.providerMetadata,
-              [state.providerMetadataKey]: { ...item.providerMetadata?.[state.providerMetadataKey], annotations },
-            },
-          })
-        : item,
-    ),
-  ] satisfies OpenResponses.StepResult
-})
-
-const step = Effect.fn("MetaResponses.step")(function* (state: ParserState, input: OpenResponses.Event) {
-  const completedItems = new Set(state.completedItems)
-  const event = OpenResponses.normalize(state, input)
-  if (event.type === "response.output_item.done" && event.item && completedItems.has(event.item.id))
-    return [state, []] as const
-  const events: LLMEvent[] = []
-  let current: OpenResponses.ParserState = state
-  // Muse Image delivers its image and optional summary only in response.completed.
-  // Recover terminal-only items in order, without duplicating Spark's streamed items.
-  if (event.type === "response.completed") {
-    for (const [index, item] of (event.response?.output ?? []).entries()) {
-      const done = OpenResponses.normalize(current, { type: "response.output_item.done", item, output_index: index })
-      // Spark changes reasoning IDs in the terminal snapshot; output indices still identify the streamed items.
-      if (!done.item || completedItems.has(done.item.id) || completedItems.has(state.outputItems[index] ?? "")) continue
-      const result = yield* onEvent(current, done)
-      current = result[0]
-      events.push(...result[1])
-      completedItems.add(done.item.id)
-    }
-  }
-  const result = yield* onEvent(current, event)
-  if (event.type === "response.output_item.done" && event.item) completedItems.add(event.item.id)
-  return [{ ...result[0], completedItems }, [...events, ...result[1]]] as const
-})
-
-export const protocol = Protocol.make({
-  id: ADAPTER,
-  body: { schema: Body, from: fromRequest },
-  stream: {
-    event: OpenResponses.protocol.stream.event,
-    initial: (request): ParserState => ({ ...OpenResponses.initial(request, adapter), completedItems: new Set() }),
-    step,
-    terminal: OpenResponses.terminal,
-  },
-})
-
-export const httpTransport = HttpTransport.sseJson.with<Schema.Schema.Type<typeof Body>>()
-
-export * as MetaResponses from "./meta-responses.js"

packages/ai/src/protocols/mistral-chat.ts

@@ -1,793 +0,0 @@
-import { Effect, Schema } from "effect"
-import { Auth } from "../route/auth.js"
-import { Route } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { Framing } from "../route/framing.js"
-import { Protocol } from "../route/protocol.js"
-import { HttpTransport } from "../route/transport/index.js"
-import {
-  AIError,
-  InvalidProviderOutputError,
-  LLMEvent,
-  ProviderInternalError,
-  UnknownProviderError,
-  Usage,
-  type FinishReasonDetails,
-  type LLMRequest,
-  type MediaPart,
-  type ToolCallPart,
-  type ToolDefinition,
-} from "../schema/index.js"
-import { classifyProviderFailure } from "../provider-error.js"
-import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
-import { Lifecycle } from "./utils/lifecycle.js"
-import { ToolStream } from "./utils/tool-stream.js"
-
-const ADAPTER = "mistral-chat"
-const DONE = "[DONE]" as const
-const TOOL_ID = /^[A-Za-z0-9]{9}$/
-export const DEFAULT_BASE_URL = "https://api.mistral.ai/v1"
-export const PATH = "/chat/completions"
-
-const MistralTextContent = Schema.Struct({
-  type: Schema.Literal("text"),
-  text: Schema.String,
-})
-
-const MistralThinkingUnit = Schema.StructWithRest(
-  Schema.Struct({
-    type: Schema.optional(Schema.String),
-    text: Schema.optional(Schema.String),
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
-type MistralThinkingUnit = Schema.Schema.Type<typeof MistralThinkingUnit>
-
-const MistralThinkingContent = Schema.StructWithRest(
-  Schema.Struct({
-    type: Schema.Literal("thinking"),
-    thinking: Schema.Array(MistralThinkingUnit),
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
-type MistralThinkingContent = Schema.Schema.Type<typeof MistralThinkingContent>
-const isMistralThinkingContent = Schema.is(MistralThinkingContent)
-
-const MistralUserContent = Schema.Union([
-  MistralTextContent,
-  Schema.Struct({ type: Schema.Literal("image_url"), image_url: Schema.String }),
-  Schema.Struct({ type: Schema.Literal("document_url"), document_url: Schema.String }),
-])
-type MistralUserContent = Schema.Schema.Type<typeof MistralUserContent>
-
-const MistralAssistantToolCall = Schema.Struct({
-  id: Schema.String,
-  type: Schema.Literal("function"),
-  function: Schema.Struct({ name: Schema.String, arguments: Schema.String }),
-})
-type MistralAssistantToolCall = Schema.Schema.Type<typeof MistralAssistantToolCall>
-
-const MistralMessage = Schema.Union([
-  Schema.Struct({ role: Schema.Literal("system"), content: Schema.String }),
-  Schema.Struct({
-    role: Schema.Literal("user"),
-    content: Schema.Union([Schema.String, Schema.Array(MistralUserContent)]),
-  }),
-  Schema.Struct({
-    role: Schema.Literal("assistant"),
-    content: Schema.Union([Schema.String, Schema.Array(Schema.Union([MistralTextContent, MistralThinkingContent]))]),
-    tool_calls: optionalArray(MistralAssistantToolCall),
-    prefix: Schema.optional(Schema.Literal(true)),
-  }),
-  Schema.Struct({
-    role: Schema.Literal("tool"),
-    tool_call_id: Schema.String,
-    name: Schema.String,
-    content: Schema.Union([Schema.String, Schema.Array(MistralUserContent)]),
-  }),
-]).pipe(Schema.toTaggedUnion("role"))
-type MistralMessage = Schema.Schema.Type<typeof MistralMessage>
-
-const MistralTool = Schema.Struct({
-  type: Schema.Literal("function"),
-  function: Schema.Struct({
-    name: Schema.String,
-    description: Schema.String,
-    parameters: JsonObject,
-    strict: Schema.Literal(false),
-  }),
-})
-type MistralTool = Schema.Schema.Type<typeof MistralTool>
-
-const MistralOptions = Schema.Struct({
-  safePrompt: Schema.optional(Schema.Boolean),
-  documentImageLimit: Schema.optional(Schema.Number),
-  documentPageLimit: Schema.optional(Schema.Number),
-  parallelToolCalls: Schema.optional(Schema.Boolean),
-  reasoningEffort: Schema.optional(Schema.String),
-  promptMode: Schema.optional(Schema.Literal("reasoning")),
-  promptCacheKey: Schema.optional(Schema.String),
-})
-
-export type ReasoningEffort = "none" | "minimal" | "low" | "medium" | "high" | "xhigh" | (string & {})
-
-export type ProviderOptionsInput = {
-  readonly safePrompt?: boolean
-  readonly documentImageLimit?: number
-  readonly documentPageLimit?: number
-  readonly parallelToolCalls?: boolean
-  readonly reasoningEffort?: ReasoningEffort
-  readonly promptMode?: "reasoning"
-  readonly promptCacheKey?: string
-  readonly [key: string]: unknown
-}
-
-const MistralBody = Schema.Struct({
-  model: Schema.String,
-  messages: Schema.Array(MistralMessage),
-  tools: optionalArray(MistralTool),
-  tool_choice: Schema.optional(
-    Schema.Union([
-      Schema.Literals(["auto", "none", "any"]),
-      Schema.Struct({ type: Schema.Literal("function"), function: Schema.Struct({ name: Schema.String }) }),
-    ]),
-  ),
-  stream: Schema.Literal(true),
-  max_tokens: Schema.optional(Schema.Number),
-  random_seed: Schema.optional(Schema.Number),
-  temperature: Schema.optional(Schema.Number),
-  top_p: Schema.optional(Schema.Number),
-  frequency_penalty: Schema.optional(Schema.Number),
-  presence_penalty: Schema.optional(Schema.Number),
-  stop: optionalArray(Schema.String),
-  prompt_cache_key: Schema.optional(Schema.String),
-  safe_prompt: Schema.optional(Schema.Boolean),
-  document_image_limit: Schema.optional(Schema.Number),
-  document_page_limit: Schema.optional(Schema.Number),
-  parallel_tool_calls: Schema.optional(Schema.Boolean),
-  reasoning_effort: Schema.optional(Schema.String),
-  prompt_mode: Schema.optional(Schema.Literal("reasoning")),
-})
-export type MistralBody = Schema.Schema.Type<typeof MistralBody>
-
-const MistralUsageDetails = Schema.StructWithRest(Schema.Struct({ cached_tokens: optionalNull(Schema.Number) }), [
-  Schema.Record(Schema.String, Schema.Unknown),
-])
-
-const MistralUsage = Schema.StructWithRest(
-  Schema.Struct({
-    prompt_tokens: optionalNull(Schema.Number),
-    completion_tokens: optionalNull(Schema.Number),
-    total_tokens: optionalNull(Schema.Number),
-    num_cached_tokens: optionalNull(Schema.Number),
-    prompt_token_details: optionalNull(MistralUsageDetails),
-    prompt_tokens_details: optionalNull(MistralUsageDetails),
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
-
-const MistralOutputContent = Schema.StructWithRest(
-  Schema.Struct({
-    type: Schema.String,
-    text: optionalNull(Schema.String),
-    thinking: optionalNull(Schema.Unknown),
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
-type MistralOutputContent = Schema.Schema.Type<typeof MistralOutputContent>
-
-const MistralToolDelta = Schema.Struct({
-  index: optionalNull(Schema.Number),
-  id: optionalNull(Schema.String),
-  function: optionalNull(
-    Schema.Struct({
-      name: optionalNull(Schema.String),
-      arguments: optionalNull(Schema.Union([Schema.String, JsonObject])),
-    }),
-  ),
-})
-type MistralToolDelta = Schema.Schema.Type<typeof MistralToolDelta>
-
-const MistralChoice = Schema.StructWithRest(
-  Schema.Struct({
-    delta: optionalNull(
-      Schema.StructWithRest(
-        Schema.Struct({
-          content: optionalNull(Schema.Union([Schema.String, Schema.Array(MistralOutputContent)])),
-          tool_calls: optionalNull(Schema.Array(MistralToolDelta)),
-        }),
-        [Schema.Record(Schema.String, Schema.Unknown)],
-      ),
-    ),
-    finish_reason: optionalNull(Schema.String),
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
-
-const MistralError = Schema.StructWithRest(
-  Schema.Struct({
-    message: Schema.String,
-    code: optionalNull(Schema.Union([Schema.String, Schema.Number])),
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
-
-const MistralEvent = Schema.StructWithRest(
-  Schema.Struct({
-    choices: optionalNull(Schema.Array(MistralChoice)),
-    usage: optionalNull(MistralUsage),
-    error: optionalNull(MistralError),
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
-type MistralEvent = Schema.Schema.Type<typeof MistralEvent>
-const MistralStreamEvent = Schema.Union([Schema.Literal(DONE), Protocol.jsonEvent(MistralEvent)])
-
-const hashID = (value: string) => {
-  const hash = (seed: number) => {
-    let result = seed
-    for (const char of value) result = Math.imul(result ^ char.charCodeAt(0), 16777619)
-    return (result >>> 0).toString(36)
-  }
-  return `${hash(2166136261).padStart(7, "0")}${hash(2246822519).padStart(7, "0")}`.slice(-9)
-}
-
-const toolIDNormalizer = (request: LLMRequest) => {
-  const ids = request.messages.flatMap((message) =>
-    message.content.flatMap((part) => (part.type === "tool-call" || part.type === "tool-result" ? [part.id] : [])),
-  )
-  const used = new Set(ids.filter((id) => TOOL_ID.test(id)))
-  const normalized = new Map<string, string>()
-  return (id: string) => {
-    if (TOOL_ID.test(id)) return id
-    const previous = normalized.get(id)
-    if (previous) return previous
-    let attempt = 0
-    let candidate = hashID(id)
-    while (used.has(candidate)) candidate = hashID(`${id}:${++attempt}`)
-    used.add(candidate)
-    normalized.set(id, candidate)
-    return candidate
-  }
-}
-
-const lowerMedia = Effect.fn("MistralChat.lowerMedia")(function* (part: MediaPart) {
-  const media = ProviderShared.normalizeMedia(part)
-  const url = typeof part.data === "string" && /^(?:https?:|data:)/.test(part.data) ? part.data : media.dataUrl
-  if (media.mime.startsWith("image/")) return { type: "image_url" as const, image_url: url }
-  if (media.mime === "application/pdf") return { type: "document_url" as const, document_url: url }
-  return yield* ProviderShared.invalidRequest(`Mistral Chat does not support media type ${part.mediaType}`)
-})
-
-const lowerUser = Effect.fn("MistralChat.lowerUser")(function* (message: LLMRequest["messages"][number]) {
-  const content: MistralUserContent[] = []
-  for (const part of message.content) {
-    if (part.type === "text") {
-      content.push({ type: "text", text: part.text })
-      continue
-    }
-    if (part.type === "media") {
-      content.push(yield* lowerMedia(part))
-      continue
-    }
-    return yield* ProviderShared.unsupportedContent("Mistral Chat", "user", ["text", "media"])
-  }
-  if (content.every((part) => part.type === "text"))
-    return { role: "user" as const, content: content.map((part) => part.text).join("") }
-  return { role: "user" as const, content }
-})
-
-const lowerToolCall = (part: ToolCallPart, normalizeID: (id: string) => string): MistralAssistantToolCall => ({
-  id: normalizeID(part.id),
-  type: "function",
-  function: { name: part.name, arguments: ProviderShared.encodeJson(part.input) },
-})
-
-const lowerAssistant = Effect.fn("MistralChat.lowerAssistant")(function* (
-  message: LLMRequest["messages"][number],
-  normalizeID: (id: string) => string,
-  prefix: boolean,
-) {
-  const structured = message.content.some(
-    (part) => part.type === "reasoning" && isMistralThinkingContent(part.providerMetadata?.mistral?.thinking),
-  )
-  const content: Array<Schema.Schema.Type<typeof MistralTextContent> | MistralThinkingContent> = []
-  const text: string[] = []
-  const toolCalls: MistralAssistantToolCall[] = []
-  for (const part of message.content) {
-    if (part.type === "text") {
-      if (structured) content.push({ type: "text", text: part.text })
-      else text.push(part.text)
-      continue
-    }
-    if (part.type === "reasoning") {
-      const native = part.providerMetadata?.mistral?.thinking
-      if (structured && isMistralThinkingContent(native)) content.push(native)
-      else if (structured) content.push({ type: "text", text: part.text })
-      else text.push(part.text)
-      continue
-    }
-    if (part.type === "tool-call") {
-      toolCalls.push(lowerToolCall(part, normalizeID))
-      continue
-    }
-    return yield* ProviderShared.unsupportedContent("Mistral Chat", "assistant", ["text", "reasoning", "tool-call"])
-  }
-  return {
-    role: "assistant" as const,
-    content: structured ? content : text.join(""),
-    ...(toolCalls.length > 0 ? { tool_calls: toolCalls } : {}),
-    ...(prefix ? { prefix: true as const } : {}),
-  }
-})
-
-const lowerToolResults = Effect.fn("MistralChat.lowerToolResults")(function* (
-  message: LLMRequest["messages"][number],
-  normalizeID: (id: string) => string,
-) {
-  const output: MistralMessage[] = []
-  for (const part of message.content) {
-    if (part.type !== "tool-result")
-      return yield* ProviderShared.unsupportedContent("Mistral Chat", "tool", ["tool-result"])
-    if (part.result.type !== "content") {
-      output.push({
-        role: "tool",
-        tool_call_id: normalizeID(part.id),
-        name: part.name,
-        content: ProviderShared.toolResultText(part),
-      })
-      continue
-    }
-    const content: MistralUserContent[] = []
-    for (const item of part.result.value) {
-      if (item.type === "text") {
-        content.push({ type: "text", text: item.text })
-        continue
-      }
-      content.push(yield* lowerMedia({ type: "media", mediaType: item.mime, data: item.uri, filename: item.name }))
-    }
-    output.push({
-      role: "tool",
-      tool_call_id: normalizeID(part.id),
-      name: part.name,
-      content: content.some((item) => item.type !== "text")
-        ? content
-        : content.map((item) => (item.type === "text" ? item.text : "")).join(""),
-    })
-  }
-  return output
-})
-
-const lowerMessages = Effect.fn("MistralChat.lowerMessages")(function* (request: LLMRequest) {
-  const normalizeID = toolIDNormalizer(request)
-  const messages: MistralMessage[] =
-    request.system.length === 0 ? [] : [{ role: "system", content: ProviderShared.joinText(request.system) }]
-  for (const message of request.messages) {
-    if (message.role === "system") {
-      const update = yield* ProviderShared.wrappedSystemUpdate("Mistral Chat", message)
-      messages.push({
-        role: "user",
-        content: update.text,
-      })
-      continue
-    }
-    if (message.role === "user") {
-      messages.push(yield* lowerUser(message))
-      continue
-    }
-    if (message.role === "assistant") {
-      const hasToolCalls = message.content.some((part) => part.type === "tool-call")
-      const hasNativeThinking = message.content.some(
-        (part) => part.type === "reasoning" && isMistralThinkingContent(part.providerMetadata?.mistral?.thinking),
-      )
-      const text = message.content
-        .flatMap((part) => (part.type === "text" || part.type === "reasoning" ? [part.text] : []))
-        .join("")
-      if (!hasToolCalls && !hasNativeThinking && text.trim() === "") continue
-      messages.push(yield* lowerAssistant(message, normalizeID, !hasToolCalls && message === request.messages.at(-1)))
-      continue
-    }
-    messages.push(...(yield* lowerToolResults(message, normalizeID)))
-  }
-  return messages
-})
-
-const lowerTool = (tool: ToolDefinition): MistralTool => ({
-  type: "function",
-  function: { name: tool.name, description: tool.description, parameters: tool.inputSchema, strict: false },
-})
-
-export const fromRequest = Effect.fn("MistralChat.fromRequest")(function* (request: LLMRequest) {
-  const options = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(MistralOptions))(
-    request.providerOptions ?? {},
-  )
-  const selected = request.toolChoice?.type === "tool" ? request.toolChoice.name : undefined
-  if (request.toolChoice?.type === "tool" && !selected)
-    return yield* ProviderShared.invalidRequest("Mistral Chat tool choice requires a tool name")
-  if (options.reasoningEffort !== undefined && options.promptMode !== undefined)
-    return yield* ProviderShared.invalidRequest(
-      "Mistral Chat reasoningEffort and promptMode provider options are mutually exclusive",
-    )
-  const toolChoice = request.toolChoice
-    ? yield* ProviderShared.matchToolChoice("Mistral Chat", request.toolChoice, {
-        auto: () => "auto" as const,
-        none: () => "none" as const,
-        required: () => "any" as const,
-        tool: (name) => ({ type: "function" as const, function: { name } }),
-      })
-    : undefined
-  const flattened = ProviderShared.flattenToolRequest(request)
-  return {
-    model: request.model.id,
-    messages: yield* lowerMessages(flattened.request),
-    tools: flattened.tools.length > 0 ? flattened.tools.map(lowerTool) : undefined,
-    tool_choice: toolChoice,
-    stream: true as const,
-    max_tokens: request.generation?.maxTokens,
-    random_seed: request.generation?.seed,
-    temperature: request.generation?.temperature,
-    top_p: request.generation?.topP,
-    frequency_penalty: request.generation?.frequencyPenalty,
-    presence_penalty: request.generation?.presencePenalty,
-    stop: request.generation?.stop,
-    prompt_cache_key: request.cache === "none" ? undefined : (options.promptCacheKey ?? request.promptCacheKey),
-    safe_prompt: options.safePrompt,
-    document_image_limit: options.documentImageLimit,
-    document_page_limit: options.documentPageLimit,
-    parallel_tool_calls:
-      options.parallelToolCalls ?? (request.toolChoice?.disableParallelToolUse === true ? false : undefined),
-    reasoning_effort: options.reasoningEffort,
-    prompt_mode: options.promptMode,
-  }
-})
-
-type ToolKey = string | number
-interface PendingTool {
-  readonly id: string
-  readonly name?: string
-  readonly input: string
-}
-
-interface ActiveContent {
-  readonly type: "text" | "reasoning"
-  readonly id: string
-  readonly thinking?: MistralThinkingContent
-}
-
-export interface ParserState {
-  readonly tools: ToolStream.State<ToolKey>
-  readonly pendingTools: Partial<Record<ToolKey, PendingTool>>
-  readonly toolIDs: ReadonlyMap<string, string>
-  readonly usedToolIDs: ReadonlySet<string>
-  readonly completedTools: ReadonlyArray<LLMEvent>
-  readonly latestToolKey?: ToolKey
-  readonly generatedTools: number
-  readonly lifecycle: Lifecycle.State
-  readonly active?: ActiveContent
-  readonly nextContent: number
-  readonly usage?: Usage
-  readonly finishReason?: FinishReasonDetails
-}
-
-const mapUsage = (usage: MistralEvent["usage"]): Usage | undefined => {
-  if (!usage) return undefined
-  const input = usage.prompt_tokens ?? undefined
-  const reported =
-    usage.num_cached_tokens ??
-    usage.prompt_tokens_details?.cached_tokens ??
-    usage.prompt_token_details?.cached_tokens ??
-    undefined
-  const cached = input === undefined || reported === undefined ? undefined : Math.max(0, Math.min(input, reported))
-  const output = usage.completion_tokens ?? undefined
-  return new Usage({
-    inputTokens: input,
-    outputTokens: output,
-    nonCachedInputTokens: ProviderShared.subtractTokens(input, cached),
-    cacheReadInputTokens: cached,
-    totalTokens: ProviderShared.totalTokens(input, output, usage.total_tokens ?? undefined),
-    providerMetadata: { mistral: usage },
-  })
-}
-
-const mapFinishReason = (reason: string) => {
-  switch (reason) {
-    case "stop":
-      return "stop" as const
-    case "length":
-    case "model_length":
-      return "length" as const
-    case "tool_calls":
-      return "tool-calls" as const
-    case "content_filter":
-      return "content-filter" as const
-    case "error":
-    case "network_error":
-      return "error" as const
-    default:
-      return "unknown" as const
-  }
-}
-
-const thinkingUnits = (value: unknown): ReadonlyArray<MistralThinkingUnit> => {
-  if (typeof value === "string") return [{ type: "text", text: value }]
-  if (!Array.isArray(value)) return []
-  return value.filter(Schema.is(MistralThinkingUnit))
-}
-
-const thinkingText = (thinking: ReadonlyArray<MistralThinkingUnit>) =>
-  thinking.flatMap((unit) => (typeof unit.text === "string" ? [unit.text] : [])).join("")
-
-const thinkingMetadata = (thinking: MistralThinkingContent) => ({ mistral: { thinking } })
-
-const closeActive = (state: ParserState, events: LLMEvent[]) => {
-  if (!state.active) return state
-  const lifecycle =
-    state.active.type === "text"
-      ? Lifecycle.textEnd(state.lifecycle, events, state.active.id)
-      : Lifecycle.reasoningEnd(
-          state.lifecycle,
-          events,
-          state.active.id,
-          thinkingMetadata(state.active.thinking ?? { type: "thinking", thinking: [] }),
-          thinkingText(state.active.thinking?.thinking ?? []),
-        )
-  return { ...state, lifecycle, active: undefined }
-}
-
-const appendText = (state: ParserState, events: LLMEvent[], text: string) => {
-  if (text.length === 0) return state
-  const current = state.active?.type === "text" ? state : closeActive(state, events)
-  const active = current.active ?? { type: "text" as const, id: `text-${current.nextContent}` }
-  return {
-    ...current,
-    lifecycle: Lifecycle.textDelta(current.lifecycle, events, active.id, text),
-    active,
-    nextContent: current.active ? current.nextContent : current.nextContent + 1,
-  }
-}
-
-const appendThinking = (state: ParserState, events: LLMEvent[], part: MistralOutputContent) => {
-  const current = state.active?.type === "reasoning" ? state : closeActive(state, events)
-  const units = thinkingUnits(part.thinking)
-  const active = current.active ?? { type: "reasoning" as const, id: `reasoning-${current.nextContent}` }
-  const thinking = {
-    ...active.thinking,
-    ...part,
-    type: "thinking" as const,
-    thinking: [...(active.thinking?.thinking ?? []), ...units],
-  }
-  const text = thinkingText(units)
-  return {
-    ...current,
-    lifecycle:
-      text.length > 0
-        ? Lifecycle.reasoningDelta(current.lifecycle, events, active.id, text, thinkingMetadata(thinking))
-        : Lifecycle.reasoningStart(current.lifecycle, events, active.id, thinkingMetadata(thinking)),
-    active: { ...active, thinking },
-    nextContent: current.active ? current.nextContent : current.nextContent + 1,
-  }
-}
-
-const appendContent = (
-  state: ParserState,
-  events: LLMEvent[],
-  content: string | ReadonlyArray<MistralOutputContent>,
-) => {
-  if (typeof content === "string") return appendText(state, events, content)
-  return content.reduce((current, part) => {
-    if (part.type === "text") return appendText(current, events, part.text ?? "")
-    if (part.type === "thinking") return appendThinking(current, events, part)
-    return closeActive(current, events)
-  }, state)
-}
-
-const normalizeStreamToolID = (state: ParserState, source: string) => {
-  if (TOOL_ID.test(source))
-    return { id: source, state: { ...state, usedToolIDs: new Set([...state.usedToolIDs, source]) } }
-  const previous = state.toolIDs.get(source)
-  if (previous) return { id: previous, state }
-  let attempt = 0
-  let id = hashID(source)
-  while (state.usedToolIDs.has(id)) id = hashID(`${source}:${++attempt}`)
-  return {
-    id,
-    state: {
-      ...state,
-      toolIDs: new Map([...state.toolIDs, [source, id]]),
-      usedToolIDs: new Set([...state.usedToolIDs, id]),
-    },
-  }
-}
-
-const toolText = (tool: MistralToolDelta) => {
-  const value = tool.function?.arguments
-  if (typeof value === "string") return value
-  return value === null || value === undefined ? "" : ProviderShared.encodeJson(value)
-}
-
-const appendTools = Effect.fn("MistralChat.appendTools")(function* (
-  initial: ParserState,
-  events: LLMEvent[],
-  deltas: ReadonlyArray<MistralToolDelta>,
-) {
-  if (deltas.length === 0) return initial
-  let state = closeActive(initial, events)
-  for (const [position, delta] of deltas.entries()) {
-    const wireID = delta.id?.trim() || undefined
-    const providedID = wireID === "null" ? undefined : wireID
-    const key =
-      delta.index ??
-      (providedID
-        ? `id:${providedID}`
-        : deltas.length > 1
-          ? `position:${position}`
-          : (state.latestToolKey ?? `missing:${state.generatedTools}`))
-    const existing = state.tools[key]
-    const pending = state.pendingTools[key]
-    const source = providedID ?? `generated:${String(key)}`
-    const normalized =
-      existing || pending ? { id: existing?.id ?? pending?.id ?? "", state } : normalizeStreamToolID(state, source)
-    state = normalized.state
-    const name = existing?.name ?? pending?.name ?? (delta.function?.name?.trim() || undefined)
-    const text = `${pending?.input ?? ""}${toolText(delta)}`
-    if (!name) {
-      state = {
-        ...state,
-        pendingTools: { ...state.pendingTools, [key]: { id: normalized.id, input: text } },
-        latestToolKey: key,
-        generatedTools: state.generatedTools + (!providedID && !pending ? 1 : 0),
-      }
-      continue
-    }
-    const result = ToolStream.appendOrStart(
-      ADAPTER,
-      state.tools,
-      key,
-      { id: normalized.id, name, text },
-      "Mistral Chat tool call delta is missing a name",
-    )
-    if (ToolStream.isError(result)) return yield* result
-    if (result.events.length > 0) state = { ...state, lifecycle: Lifecycle.stepStart(state.lifecycle, events) }
-    events.push(...result.events)
-    const pendingTools = { ...state.pendingTools }
-    delete pendingTools[key]
-    state = {
-      ...state,
-      tools: result.tools,
-      pendingTools,
-      latestToolKey: key,
-      generatedTools: state.generatedTools + (!providedID && !existing && !pending ? 1 : 0),
-    }
-  }
-  return state
-})
-
-const hasLateContent = (event: MistralEvent) => {
-  const delta = event.choices?.[0]?.delta
-  if (typeof delta?.content === "string" && delta.content.length > 0) return true
-  if (Array.isArray(delta?.content) && delta.content.length > 0) return true
-  return (delta?.tool_calls ?? []).some(
-    (tool) => Boolean(tool.id) || Boolean(tool.function?.name) || tool.function?.arguments !== undefined,
-  )
-}
-
-const step = Effect.fn("MistralChat.step")(function* (state: ParserState, event: MistralEvent) {
-  if (event.error) {
-    const body = ProviderShared.encodeJson(event)
-    return yield* new AIError({
-      reason: classifyProviderFailure({
-        message: event.error.message,
-        status: typeof event.error.code === "number" ? event.error.code : undefined,
-        rawBody: body,
-      }),
-    })
-  }
-  const events: LLMEvent[] = []
-  const usage = mapUsage(event.usage) ?? state.usage
-  if (state.finishReason) {
-    if (hasLateContent(event))
-      return yield* ProviderShared.eventError(
-        ADAPTER,
-        "Mistral Chat received content after the finish reason",
-        ProviderShared.encodeJson(event),
-      )
-    return [{ ...state, usage }, events] as const
-  }
-  const choice = event.choices?.[0]
-  const withContent = choice?.delta?.content == null ? state : appendContent(state, events, choice.delta.content)
-  const withTools = yield* appendTools(withContent, events, choice?.delta?.tool_calls ?? [])
-  if (!choice?.finish_reason) return [{ ...withTools, usage }, events] as const
-
-  const finishReason = {
-    normalized: mapFinishReason(choice.finish_reason),
-    raw: choice.finish_reason,
-  }
-  if (finishReason.normalized === "error") {
-    const details = {
-      message: `Mistral Chat stopped with ${finishReason.raw}`,
-      body: ProviderShared.encodeJson(event),
-    }
-    return yield* new AIError({
-      reason:
-        finishReason.raw === "network_error" ? new ProviderInternalError(details) : new UnknownProviderError(details),
-    })
-  }
-  const incomplete = finishReason.normalized === "length" || finishReason.normalized === "content-filter"
-  if (!incomplete && Object.keys(withTools.pendingTools).length > 0)
-    return yield* ProviderShared.eventError(
-      ADAPTER,
-      "Mistral Chat tool call delta is missing a name",
-      ProviderShared.encodeJson(event),
-    )
-  const finished =
-    !incomplete && Object.keys(withTools.tools).length > 0
-      ? yield* ToolStream.finishAll(ADAPTER, withTools.tools)
-      : undefined
-  return [
-    {
-      ...withTools,
-      tools: finished?.tools ?? withTools.tools,
-      completedTools: finished?.events ?? withTools.completedTools,
-      usage,
-      finishReason,
-    },
-    events,
-  ] as const
-})
-
-const finishEvents = Effect.fn("MistralChat.finishEvents")(function* (state: ParserState) {
-  if (!state.finishReason)
-    return yield* new AIError({
-      reason: new InvalidProviderOutputError({
-        message: "Mistral Chat stream ended without finish_reason",
-        classification: "incomplete-stream",
-        route: ADAPTER,
-      }),
-    })
-  const events: LLMEvent[] = []
-  const closed = closeActive(state, events)
-  const lifecycle = closed.completedTools.length > 0 ? Lifecycle.stepStart(closed.lifecycle, events) : closed.lifecycle
-  events.push(...closed.completedTools)
-  const reason =
-    state.finishReason.normalized === "stop" && closed.completedTools.some(LLMEvent.is.toolCall)
-      ? { ...state.finishReason, normalized: "tool-calls" as const }
-      : state.finishReason
-  Lifecycle.finish(lifecycle, events, { reason, usage: closed.usage })
-  return events
-})
-
-export const protocol = Protocol.make({
-  id: ADAPTER,
-  body: { schema: MistralBody, from: fromRequest },
-  stream: {
-    event: MistralStreamEvent,
-    initial: (): ParserState => ({
-      tools: ToolStream.empty<ToolKey>(),
-      pendingTools: {},
-      toolIDs: new Map(),
-      usedToolIDs: new Set(),
-      completedTools: [],
-      generatedTools: 0,
-      lifecycle: Lifecycle.initial(),
-      nextContent: 0,
-    }),
-    step: (state: ParserState, event) => (event === DONE ? Effect.succeed([state, []] as const) : step(state, event)),
-    terminal: (event) => event === DONE,
-    onHalt: finishEvents,
-  },
-})
-
-export const framing = Framing.sseWithDone
-export const httpTransport = HttpTransport.sseJson.with<MistralBody>().with({ framing })
-
-export const route = Route.make({
-  id: ADAPTER,
-  provider: "mistral",
-  providerMetadataKey: "mistral",
-  protocol,
-  endpoint: Endpoint.path(PATH, { baseURL: DEFAULT_BASE_URL }),
-  auth: Auth.none,
-  transport: httpTransport,
-})
-
-export * as MistralChat from "./mistral-chat.js"

packages/ai/src/protocols/open-responses-channel.ts

@@ -113,8 +113,8 @@ const driver = (options: Options, body: string): WebSocketChannelDriver => {
           responseID = created
           return { type: "frame", frame }
         }
-        // Keepalives and provider notifications carry no response state and may precede response.created.
-        if (!event.type.startsWith("response.")) return { type: "frame", frame }
+        // Keepalives carry no response state and may arrive before response.created.
+        if (event.type === "keepalive") return { type: "frame", frame }
         if (!responseID)
           return yield* ProviderShared.eventError(
             options.id,

packages/ai/src/protocols/open-responses-continuation.ts

@@ -42,7 +42,6 @@ const canonical = (value: unknown): string => {
   if (Array.isArray(value)) return `[${value.map(canonical).join(",")}]`
   if (!ProviderShared.isRecord(value)) return ProviderShared.encodeJson(value)
   return `{${Object.keys(value)
-    .filter((key) => value[key] !== undefined)
     .sort()
     .map((key) => `${ProviderShared.encodeJson(key)}:${canonical(value[key])}`)
     .join(",")}}`
@@ -58,12 +57,7 @@ const comparable = (value: unknown) => {
   if (value.type === "message" && value.role === "assistant")
     return {
       role: "assistant",
-      // Annotations and logprobs describe the response, not the text replayed in model input.
-      content: Array.isArray(value.content)
-        ? value.content.map((part) =>
-            ProviderShared.isRecord(part) && part.type === "output_text" ? { type: part.type, text: part.text } : part,
-          )
-        : value.content,
+      content: value.content,
       ...(value.phase === undefined ? {} : { phase: value.phase }),
     }
   if (value.type === "function_call")
@@ -127,7 +121,7 @@ const rejected = (
 
 export const driver = (input: DriverInput): WebSocketChannelDriver => {
   const { previous_response_id: _previousResponseID, ...request } = input.request
-  let output: OpenResponses.StreamItem[] = []
+  let output: unknown[] = []
   return {
     create: (checkpoint) =>
       Effect.sync(() => {
@@ -155,12 +149,6 @@ export const driver = (input: DriverInput): WebSocketChannelDriver => {
           if (rejection === "websocket_connection_limit_reached") return rejected(observation, "rotate-and-retry-full")
         }
         if (observation.type !== "completed") return observation
-        // A trigger installs a different context window. Clear the append baseline, retaining the socket.
-        if (
-          Array.isArray(request.input) &&
-          request.input.some((item) => ProviderShared.isRecord(item) && item.type === "compaction_trigger")
-        )
-          return observation
         const responseID = event.response?.id
         if (!responseID || responseID.trim().length === 0) return observation
         return {
@@ -171,14 +159,7 @@ export const driver = (input: DriverInput): WebSocketChannelDriver => {
               version: VERSION,
               responseID,
               request,
-              // Completion can re-encrypt reasoning. Callers replay the item already emitted by output_item.done.
-              output: event.response?.output
-                ? event.response.output.map((item) =>
-                    item.type === "reasoning" && item.id !== undefined
-                      ? (output.find((done) => done.type === item.type && done.id === item.id) ?? item)
-                      : item,
-                  )
-                : output.slice(),
+              output: event.response?.output ? [...event.response.output] : output.slice(),
             } satisfies CheckpointValue,
           },
         }

packages/ai/src/protocols/openai-chat.ts

@@ -1,9 +1,8 @@
 import { Effect, Schema } from "effect"
-import { Tool } from "@opencode/schema/tool"
+import { Tool } from "@opencode-ai/schema/tool"
 import { Route } from "../route/client.js"
 import { Auth } from "../route/auth.js"
 import { Endpoint } from "../route/endpoint.js"
-import { Framing } from "../route/framing.js"
 import { HttpTransport } from "../route/transport/index.js"
 import { Protocol } from "../route/protocol.js"
 import {
@@ -246,8 +245,6 @@ export const OpenAIChatEvent = Schema.StructWithRest(
   [Schema.Record(Schema.String, Schema.Unknown)],
 )
 export type OpenAIChatEvent = Schema.Schema.Type<typeof OpenAIChatEvent>
-const DONE = "[DONE]" as const
-const OpenAIChatStreamEvent = Schema.Union([Schema.Literal(DONE), Protocol.jsonEvent(OpenAIChatEvent)])
 type OpenAIChatRequestMessage = LLMRequest["messages"][number]
 
 interface PendingToolDelta {
@@ -736,7 +733,6 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
     )
   const generation = request.generation
   const toolSchemaCompatibility = request.model.compatibility?.toolSchema
-  const flattened = ProviderShared.flattenToolRequest(request)
   const provider = String(request.model.provider)
   const baseURL = request.model.route.endpoint.baseURL
   const detectedMaxTokensField = detectMaxTokensField(provider, baseURL)
@@ -749,16 +745,16 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
   const zaiToolStream =
     request.model.compatibility?.zaiToolStream ?? detectZaiToolStream(provider, baseURL, request.model.id)
   const hasHistory = hasToolHistory(request.messages)
-  const hasActiveTools = flattened.tools.length > 0
+  const hasActiveTools = request.tools.length > 0
   return {
     model: request.model.id,
-    messages: yield* lowerMessages(flattened.request, options),
+    messages: yield* lowerMessages(request, options),
     tools:
-      flattened.tools.length === 0
+      request.tools.length === 0
         ? hasHistory
           ? []
           : undefined
-        : flattened.tools.map((tool) =>
+        : request.tools.map((tool) =>
             lowerTool(
               tool,
               ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility),
@@ -816,12 +812,7 @@ const mapFinishReason = Effect.fn("OpenAIChat.mapFinishReason")(function* (event
     case "tool_calls":
       return "tool-calls" as const
     default:
-      return yield* new AIError({
-        reason: new UnknownProviderError({
-          message: `Provider finish_reason: ${reason}`,
-          body: ProviderShared.encodeJson(event),
-        }),
-      })
+      return "unknown" as const
   }
 })
 
@@ -1006,12 +997,33 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
       lifecycle = Lifecycle.reasoningStart(lifecycle, events, "reasoning-0", deltaMetadata)
     const reasoningEmitted = state.reasoningEmitted || lifecycle.reasoning.has("reasoning-0")
 
-    // Reasoning is one response-wide channel: it stays open alongside text and
-    // refusal output so late reasoning deltas and details join the same block,
-    // and `finishEvents` closes it once with the complete metadata.
-    if (delta?.content) lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)
+    if (delta?.content) {
+      lifecycle = Lifecycle.reasoningEnd(
+        lifecycle,
+        events,
+        "reasoning-0",
+        reasoningMetadata(
+          state.providerMetadataKey,
+          reasoningField,
+          reasoningDetailsObserved ? state.reasoningDetails : undefined,
+        ),
+      )
+      lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)
+    }
 
-    if (delta?.refusal) lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.refusal)
+    if (delta?.refusal) {
+      lifecycle = Lifecycle.reasoningEnd(
+        lifecycle,
+        events,
+        "reasoning-0",
+        reasoningMetadata(
+          state.providerMetadataKey,
+          reasoningField,
+          reasoningDetailsObserved ? state.reasoningDetails : undefined,
+        ),
+      )
+      lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.refusal)
+    }
 
     // Compatible providers may omit indexes. Prefer durable identity, then use
     // batch position for parallel deltas or the latest call for sparse chunks.
@@ -1060,25 +1072,17 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
       events.push(...result.events)
     }
 
-    const incompleteTools = finishReason?.normalized === "content-filter" || finishReason?.normalized === "length"
-    if (
-      finishReason !== undefined &&
-      !incompleteTools &&
-      state.finishReason === undefined &&
-      Object.keys(pendingTools).length
-    )
+    if (finishReason !== undefined && state.finishReason === undefined && Object.keys(pendingTools).length > 0)
       return yield* ProviderShared.eventError(
         ADAPTER,
         "OpenAI Chat tool call delta is missing id or name",
         ProviderShared.encodeJson(event),
       )
 
-    // Filtering or truncation terminates the response without confirming pending tool calls.
+    // Finalize accumulated tool inputs eagerly when finish_reason arrives so
+    // valid calls and malformed local calls settle independently.
     const finished =
-      finishReason !== undefined &&
-      !incompleteTools &&
-      state.finishReason === undefined &&
-      Object.keys(tools).length > 0
+      finishReason !== undefined && state.finishReason === undefined && Object.keys(tools).length > 0
         ? yield* ToolStream.finishAll(ADAPTER, tools)
         : undefined
 
@@ -1125,12 +1129,10 @@ const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: Pars
           state.finishReason.normalized === "stop" && hasToolCalls ? "tool-calls" : state.finishReason.normalized,
       }
     : { normalized: hasToolCalls ? ("tool-calls" as const) : ("stop" as const) }
-  // Snapshot details at publish time so the emitted event never observes later
-  // mutation of the accumulated `reasoningDetails` array.
   const metadata = reasoningMetadata(
     state.providerMetadataKey,
     state.reasoningField,
-    state.reasoningDetailsObserved ? [...state.reasoningDetails] : undefined,
+    state.reasoningDetailsObserved ? state.reasoningDetails : undefined,
   )
   const started =
     state.reasoningDetailsObserved && !state.reasoningEmitted
@@ -1164,7 +1166,7 @@ export const protocol = Protocol.make({
     from: fromRequest,
   },
   stream: {
-    event: OpenAIChatStreamEvent,
+    event: Protocol.jsonEvent(OpenAIChatEvent),
     initial: (request) => ({
       providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
       tools: ToolStream.empty<number>(),
@@ -1178,14 +1180,12 @@ export const protocol = Protocol.make({
       nextToolIndex: 0,
       requireFinishReason: request.model.compatibility?.requireFinishReason ?? true,
     }),
-    step: (state: ParserState, event) => (event === DONE ? Effect.succeed([state, []] as const) : step(state, event)),
-    terminal: (event) => event === DONE,
+    step,
     onHalt: finishEvents,
   },
 })
 
-export const framing = Framing.sseWithDone
-export const httpTransport = HttpTransport.sseJson.with<OpenAIChatBody>().with({ framing })
+export const httpTransport = HttpTransport.sseJson.with<OpenAIChatBody>()
 
 export const route = Route.make({
   id: ADAPTER,

packages/ai/src/protocols/openai-compatible-chat.ts

@@ -1,5 +1,6 @@
 import { Route, type RouteRoutedLanguageModelInput } from "../route/client.js"
 import { Endpoint } from "../route/endpoint.js"
+import { Framing } from "../route/framing.js"
 import * as OpenAIChat from "./openai-chat.js"
 
 const ADAPTER = "openai-compatible-chat"
@@ -18,7 +19,7 @@ export const route = Route.make({
   providerMetadataKey: "openai",
   protocol: OpenAIChat.protocol,
   endpoint: Endpoint.path("/chat/completions"),
-  framing: OpenAIChat.framing,
+  framing: Framing.sse,
 })
 
 export * as OpenAICompatibleChat from "./openai-compatible-chat.js"

packages/ai/src/protocols/openai-responses.ts

@@ -5,15 +5,13 @@ import { Auth } from "../route/auth.js"
 import { Endpoint } from "../route/endpoint.js"
 import { Protocol } from "../route/protocol.js"
 import { HttpTransport } from "../route/transport/index.js"
-import { LLMRequest, mergeJsonRecords, type JsonSchema, type ToolDefinition, type ToolEntry } from "../schema/index.js"
+import { LLMRequest, type JsonSchema, type ToolDefinition } from "../schema/index.js"
 import { OpenResponses } from "./open-responses.js"
 import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
 import { OpenAIImage } from "./utils/openai-image.js"
 import { ResponsesHostedTools } from "./utils/responses-hosted-tools.js"
 import { ToolSchemaProjection } from "./utils/tool-schema.js"
 import { OpenResponsesChannel } from "./open-responses-channel.js"
-import { ResponsesCompaction } from "./utils/responses-compaction.js"
-import { ResponsesCheckpoint } from "./utils/responses-checkpoint.js"
 
 const ADAPTER = "openai-responses"
 const NAME = "OpenAI Responses"
@@ -22,14 +20,6 @@ const WEBSOCKET_ROTATE_AFTER_MS = 55 * 60 * 1000
 export const DEFAULT_BASE_URL = "https://api.openai.com/v1"
 export const PATH = OpenResponses.PATH
 
-export const ContextManagement = Schema.Array(
-  Schema.Struct({
-    type: Schema.Literal("compaction"),
-    compactThreshold: Schema.optional(Schema.Int.check(Schema.isGreaterThan(0))),
-  }),
-)
-export type ContextManagement = typeof ContextManagement.Type
-
 const OpenAIResponsesImageGenerationTool = Schema.Struct({
   type: Schema.tag("image_generation"),
   action: Schema.optional(Schema.Literals(["auto", "generate", "edit"])),
@@ -76,18 +66,7 @@ const OpenAIResponsesHostedToolItem = Schema.Union([
   ),
 ])
 
-const OpenAIResponsesNamespace = Schema.Struct({
-  type: Schema.tag("namespace"),
-  name: Schema.String,
-  description: Schema.String,
-  tools: Schema.Array(OpenResponses.Tool),
-})
-
-const OpenAIResponsesTools = Schema.Union([
-  OpenResponses.Tool,
-  OpenAIResponsesNamespace,
-  OpenAIResponsesImageGenerationTool,
-])
+const OpenAIResponsesTools = Schema.Union([OpenResponses.Tool, OpenAIResponsesImageGenerationTool])
 
 const OpenAIResponsesToolChoice = Schema.Union([
   OpenResponses.ToolChoice,
@@ -99,14 +78,6 @@ const OpenAIResponsesCoreFields = {
   input: Schema.Array(Schema.Union([OpenResponses.InputItem, OpenAIResponsesHostedToolItem])),
   tools: optionalArray(OpenAIResponsesTools),
   tool_choice: Schema.optional(OpenAIResponsesToolChoice),
-  context_management: Schema.optional(
-    Schema.Array(
-      Schema.Struct({
-        type: Schema.Literal("compaction"),
-        compact_threshold: Schema.optional(Schema.Int.check(Schema.isGreaterThan(0))),
-      }),
-    ),
-  ),
 }
 
 const OpenAIResponsesBody = Schema.Struct({
@@ -115,23 +86,11 @@ const OpenAIResponsesBody = Schema.Struct({
 })
 export type OpenAIResponsesBody = Schema.Schema.Type<typeof OpenAIResponsesBody>
 
-/** Request control, never conversation content. */
-export const CompactionTrigger = Schema.Struct({ type: Schema.Literal("compaction_trigger") })
-const CheckpointBody = Schema.Struct({
-  ...OpenAIResponsesBody.fields,
-  input: Schema.Array(Schema.Union([OpenResponses.InputItem, OpenAIResponsesHostedToolItem, CompactionTrigger])),
-  store: Schema.Literal(false),
-  prompt_cache_retention: optionalNull(Schema.String),
-  prompt_cache_options: optionalNull(
-    Schema.Struct({ mode: Schema.optional(Schema.String), ttl: Schema.optional(Schema.String) }),
-  ),
-})
-
-const adapter = {
+const extension = {
   id: ADAPTER,
   name: NAME,
-  restoreHostedToolItem: (item: unknown) => (Schema.is(OpenAIResponsesHostedToolItem)(item) ? item : undefined),
-} satisfies OpenResponses.ProviderAdapter
+  lowerHostedToolItem: (item: unknown) => (Schema.is(OpenAIResponsesHostedToolItem)(item) ? item : undefined),
+} satisfies OpenResponses.Extension
 
 const nativeImageToolInput = (tool: ToolDefinition) => {
   const native = tool.native?.openai
@@ -152,33 +111,13 @@ const lowerTool = Effect.fn("OpenAIResponses.lowerTool")(function* (tool: ToolDe
   return yield* OpenResponses.lowerTool(NAME, tool, inputSchema)
 })
 
-// Native namespaces hold only function tools, so deeper levels flatten into
-// the leaf names the same way non-native protocols flatten the whole tree.
-const lowerToolEntry = Effect.fn("OpenAIResponses.lowerToolEntry")(function* (
-  tool: ToolEntry,
-  compatibility: Parameters<typeof ToolSchemaProjection.modelCompatibility>[1],
-) {
-  if (tool.type === "tool")
-    return yield* lowerTool(tool, ToolSchemaProjection.modelCompatibility(tool.inputSchema, compatibility))
-  // OpenAI requires a namespace description; fall back to a generic one so a
-  // missing description never blocks the request.
-  return {
-    type: "namespace" as const,
-    name: tool.name,
-    description: tool.description ?? `Tools in the ${tool.name} namespace.`,
-    tools: yield* Effect.forEach(ProviderShared.flattenTools(tool.tools), (leaf) =>
-      OpenResponses.lowerTool(NAME, leaf, ToolSchemaProjection.modelCompatibility(leaf.inputSchema, compatibility)),
-    ),
-  }
-})
-
-const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>, tools: ReadonlyArray<ToolEntry>) =>
+const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>, tools: ReadonlyArray<ToolDefinition>) =>
   ProviderShared.matchToolChoice(NAME, toolChoice, {
     auto: () => "auto" as const,
     none: () => "none" as const,
     required: () => "required" as const,
     tool: (name) =>
-      tools.some((tool) => tool.type === "tool" && tool.name === name && nativeImageTool(tool) !== undefined)
+      tools.some((tool) => tool.name === name && nativeImageTool(tool) !== undefined)
         ? ({ type: "image_generation" } as const)
         : { type: "function" as const, name },
   })
@@ -186,53 +125,26 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>, tool
 const decodeBody = ProviderShared.validateWith(Schema.decodeUnknownEffect(OpenAIResponsesBody))
 
 const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request: LLMRequest) {
-  const management = yield* ProviderShared.validateWith(
-    Schema.decodeUnknownEffect(Schema.UndefinedOr(ContextManagement)),
-  )(request.providerOptions?.contextManagement)
+  const body = yield* OpenResponses.fromRequestWithExtension(
+    LLMRequest.update(request, { tools: [], toolChoice: undefined }),
+    extension,
+  )
   const toolSchemaCompatibility = request.model.compatibility?.toolSchema
+  const parallelToolCalls = OpenResponses.resolveParallelToolCalls(request)
   return yield* decodeBody({
-    ...(yield* OpenResponses.lowerConversation(request, adapter)),
-    ...OpenResponses.lowerGeneration(request),
-    context_management: management?.map((edit) => ({ type: edit.type, compact_threshold: edit.compactThreshold })),
+    ...body,
+    ...(parallelToolCalls === undefined ? {} : { parallel_tool_calls: parallelToolCalls }),
     tools:
       request.tools.length === 0
         ? undefined
-        : yield* Effect.forEach(request.tools, (tool) => lowerToolEntry(tool, toolSchemaCompatibility)),
+        : yield* Effect.forEach(request.tools, (tool) =>
+            lowerTool(tool, ToolSchemaProjection.modelCompatibility(tool.inputSchema, toolSchemaCompatibility)),
+          ),
     tool_choice:
-      OpenResponses.allowedToolChoice(request) ??
-      (request.toolChoice ? yield* lowerToolChoice(request.toolChoice, request.tools) : undefined),
+      body.tool_choice ?? (request.toolChoice ? yield* lowerToolChoice(request.toolChoice, request.tools) : undefined),
   })
 })
 
-const checkpointBody = {
-  schema: CheckpointBody,
-  from: Effect.fn("OpenAIResponses.checkpointBody")(function* (request: LLMRequest) {
-    const native = yield* fromRequest(LLMRequest.update(request, { toolChoice: undefined }))
-    const overlay = request.http?.body
-    // Complete history is required for stateless replay and SSE recovery. Raw input overrides bypass that contract.
-    if (
-      overlay?.input !== undefined ||
-      overlay?.previous_response_id !== undefined ||
-      overlay?.conversation !== undefined
-    )
-      return yield* ProviderShared.invalidRequest(
-        "Trigger compaction requires complete canonical history, not an input or continuation override",
-      )
-    return yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(CheckpointBody))({
-      ...mergeJsonRecords(native, overlay),
-      input: [...native.input, { type: "compaction_trigger" }],
-      stream: true,
-      store: false,
-      parallel_tool_calls: true,
-      tool_choice: undefined,
-      context_management: undefined,
-      text: undefined,
-      max_output_tokens: undefined,
-      max_tool_calls: undefined,
-    })
-  }),
-}
-
 const hostedToolResult = Effect.fn("OpenAIResponses.hostedToolResult")(function* (item: ResponsesHostedTools.Item) {
   const isError = item.error !== undefined && item.error !== null
   if (item.type === "image_generation_call" && item.result) {
@@ -272,11 +184,12 @@ const HOSTED_TOOLS = {
   },
 } as const satisfies ResponsesHostedTools.Definitions
 
-const step = (state: OpenResponses.ParserState, input: OpenResponses.Event) => {
-  const event = OpenResponses.normalize(state, input)
+const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {
   if (event.type === "response.reasoning_text.delta")
     return event.item_id !== undefined
-      ? Effect.succeed(OpenResponses.onReasoningDelta(state, event, event.item_id))
+      ? Effect.succeed(
+          OpenResponses.onReasoningDelta(state, event, OpenResponses.outputItemID(state, event) ?? event.item_id),
+        )
       : ProviderShared.eventError(ADAPTER, `${event.type} is missing item_id`)
   if (event.type === "response.output_item.done" && event.item && ResponsesHostedTools.isItem(event.item, HOSTED_TOOLS))
     return ResponsesHostedTools.onDone(state, event.item, HOSTED_TOOLS)
@@ -291,7 +204,7 @@ export const protocol = Protocol.make({
   },
   stream: {
     event: OpenResponses.protocol.stream.event,
-    initial: (request) => OpenResponses.initial(request, adapter),
+    initial: (request) => OpenResponses.initial(request, extension),
     step,
     terminal: OpenResponses.terminal,
   },
@@ -310,7 +223,6 @@ export const transport = channelTransport({
 })
 
 export const route = Route.make({
-  compact: { endpoint: ResponsesCompaction.make(adapter), trigger: ResponsesCheckpoint.make(checkpointBody) },
   id: ADAPTER,
   provider: "openai",
   providerMetadataKey: "openai",

packages/ai/src/protocols/shared.ts

@@ -1,22 +1,17 @@
 import { Buffer } from "node:buffer"
-import { Tool } from "@opencode/schema/tool"
+import { Tool } from "@opencode-ai/schema/tool"
 import { Effect, Schema, Stream } from "effect"
 import * as Sse from "effect/unstable/encoding/Sse"
 import { Headers, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
 import {
   InvalidProviderOutputError,
   InvalidRequestError,
-  UnsupportedOperationError,
   AIError,
   HttpContext,
-  LLMRequest,
-  Message,
-  ToolDefinition,
   type ContentPart,
+  type LLMRequest,
   type MediaPart,
-  type ProviderID,
   type TextPart,
-  type ToolEntry,
   type ToolResultPart,
 } from "../schema/index.js"
 import { isRecord } from "../utils/record.js"
@@ -49,7 +44,6 @@ export const promptCacheKey = (request: LLMRequest): string | undefined => {
 export interface ToolAccumulator {
   readonly id: string
   readonly name: string
-  readonly namespace?: string
   readonly input: string
 }
 
@@ -213,16 +207,15 @@ export const errorText = (error: unknown) => {
 
 /**
  * `framing` step for Server-Sent Events. Decodes UTF-8, runs the SSE channel
- * decoder, optionally filters named events, and drops empty events. `[DONE]`
- * is dropped by default or retained for protocols that use it as their stream
- * boundary. Retry control events are ignored without interrupting the stream.
+ * decoder, optionally filters named events, and drops empty / `[DONE]`
+ * keep-alive events so the protocol event schema sees one JSON string per
+ * element. Retry control events are ignored without interrupting the stream.
  * Decoder failures become provider output errors so the public error channel
  * stays `AIError`.
  */
 export const sseFraming = (
   bytes: Stream.Stream<Uint8Array, AIError>,
   events?: ReadonlySet<string>,
-  includeDone = false,
 ): Stream.Stream<string, AIError> =>
   bytes.pipe(
     Stream.decodeText(),
@@ -247,7 +240,7 @@ export const sseFraming = (
       (event) =>
         (events === undefined || events.has(event.event)) &&
         event.data.length > 0 &&
-        (event.data !== "[DONE]" || includeDone || (events !== undefined && event.event !== "message")),
+        (event.data !== "[DONE]" || (events !== undefined && event.event !== "message")),
     ),
     Stream.map((event) => event.data),
   )
@@ -260,61 +253,6 @@ export const invalidRequest = (message: string, cause?: unknown) =>
     reason: new InvalidRequestError({ message, cause }),
   })
 
-/**
- * Canonical constructor for operations the selected route does not implement.
- * Prefer this over `invalidRequest` when the failure is a missing route
- * capability rather than a malformed caller input, so consumers can branch on
- * `reason._tag` plus `reason.operation` instead of matching message text.
- */
-export const unsupportedOperation = (input: {
-  readonly operation: string
-  readonly message: string
-  readonly provider?: ProviderID
-  readonly route?: string
-  readonly cause?: unknown
-}) =>
-  new AIError({
-    reason: new UnsupportedOperationError({
-      operation: input.operation,
-      message: input.message,
-      provider: input.provider,
-      route: input.route,
-      cause: input.cause,
-    }),
-  })
-
-/**
- * Lower namespaces to flat definitions for protocols without a native
- * namespace construct. Leaf names join their namespace path with `_` because
- * `.` is not broadly accepted in provider tool names.
- */
-export const flattenTools = (tools: ReadonlyArray<ToolEntry>, path: ReadonlyArray<string> = []) => {
-  const flat = tools.flatMap((tool): ReadonlyArray<ToolDefinition> => {
-    if (tool.type === "namespace") return flattenTools(tool.tools, [...path, tool.name])
-    if (path.length === 0) return [tool]
-    return [new ToolDefinition({ ...tool, name: [...path, tool.name].join("_") })]
-  })
-  return [...new Map(flat.map((tool) => [tool.name, tool])).values()]
-}
-
-export const flattenToolRequest = (request: LLMRequest) => {
-  const messages = request.messages.map((message) => {
-    const content = message.content.map((part) => {
-      if ((part.type !== "tool-call" && part.type !== "tool-result") || part.namespace === undefined) return part
-      return { ...part, name: `${part.namespace}_${part.name}`, namespace: undefined }
-    })
-    return content.every((part, index) => part === message.content[index])
-      ? message
-      : new Message({ ...message, content })
-  })
-  return {
-    tools: flattenTools(request.tools),
-    request: messages.every((message, index) => message === request.messages[index])
-      ? request
-      : LLMRequest.update(request, { messages }),
-  }
-}
-
 export const imageResponse = Effect.fn("ProviderShared.imageResponse")(function* (
   route: string,
   name: string,

packages/ai/src/protocols/utils/bedrock-auth.ts

@@ -2,12 +2,13 @@ import { AwsV4Signer } from "aws4fetch"
 import { Effect } from "effect"
 import { Headers } from "effect/unstable/http"
 import { Auth, type AuthInput } from "../../route/auth.js"
-import { AIError, AuthenticationError } from "../../schema/index.js"
 import { ProviderShared } from "../shared.js"
 
 /**
  * AWS credentials for SigV4 signing. Bedrock also supports Bearer API key auth,
- * which provider facades configure as route auth instead of SigV4.
+ * which provider facades configure as route auth instead of SigV4. STS-vended
+ * credentials should be refreshed by the consumer (rebuild the model) before
+ * they expire; the route does not refresh.
  */
 export interface Credentials {
   readonly region: string
@@ -16,44 +17,6 @@ export interface Credentials {
   readonly sessionToken?: string
 }
 
-/** Static credentials or an effect resolved before every request. */
-export type CredentialSource = Credentials | Effect.Effect<Credentials, AIError>
-
-export interface DefaultChainOptions {
-  readonly region: string
-  /** Shared config profile passed to the AWS default chain. */
-  readonly profile?: string
-}
-
-/**
- * Resolve credentials through the AWS default provider chain: environment
- * variables, shared config and SSO caches, web identity tokens, process
- * credentials, and container or instance metadata. A fresh chain runs on every
- * request so credentials rotated on disk without an expiration (for example
- * shared-config keys rewritten by a corporate SSO tool) are always re-read;
- * the SDK's own memoization would otherwise pin them for the process lifetime.
- */
-export const defaultChain = (options: DefaultChainOptions): Effect.Effect<Credentials, AIError> =>
-  Effect.tryPromise({
-    try: async () => {
-      const { fromNodeProviderChain } = await import("@aws-sdk/credential-providers")
-      const identity = await fromNodeProviderChain(options.profile === undefined ? {} : { profile: options.profile })()
-      return {
-        region: options.region,
-        accessKeyId: identity.accessKeyId,
-        secretAccessKey: identity.secretAccessKey,
-        ...(identity.sessionToken === undefined ? {} : { sessionToken: identity.sessionToken }),
-      }
-    },
-    catch: (error) =>
-      new AIError({
-        reason: new AuthenticationError({
-          message: `AWS default credential chain failed: ${ProviderShared.errorText(error)}`,
-          cause: error,
-        }),
-      }),
-  })
-
 const signRequest = (input: {
   readonly url: string
   readonly body: string
@@ -85,17 +48,16 @@ const signRequest = (input: {
 
 /** Sign the exact JSON bytes with SigV4 using credentials configured on the route. */
 export const sigV4 = (
-  source: CredentialSource | undefined,
+  credentials: Credentials | undefined,
   options: { readonly service?: string; readonly name?: string } = {},
 ) =>
   Auth.custom((input: AuthInput) => {
     return Effect.gen(function* () {
-      if (!source) {
+      if (!credentials) {
         return yield* ProviderShared.invalidRequest(
           `${options.name ?? "Bedrock Converse"} requires either route bearer auth or AWS credentials configured on the route`,
         )
       }
-      const credentials = Effect.isEffect(source) ? yield* source : source
       const headersForSigning = Headers.set(input.headers, "content-type", "application/json")
       const signed = yield* signRequest({
         url: input.url,
@@ -112,35 +74,4 @@ export const sigV4 = (
 /** Bedrock route auth defaults to SigV4 and expects credentials from route configuration. */
 export const auth = sigV4(undefined)
 
-export const resolveRegion = (input: {
-  readonly region?: string
-  readonly credentials?: { readonly region: string }
-}) =>
-  input.region ?? input.credentials?.region ?? process.env.AWS_REGION ?? process.env.AWS_DEFAULT_REGION ?? "us-east-1"
-
-export interface ResolveAuthInput {
-  readonly apiKey?: string
-  readonly credentials?: Credentials
-  readonly profile?: string
-}
-
-export interface ResolveAuthOptions {
-  readonly service?: string
-  readonly name?: string
-  /** `sigv4` ignores an ambient `AWS_BEARER_TOKEN_BEDROCK`; `bearer` is validated by the caller. */
-  readonly mode?: "bearer" | "sigv4"
-}
-
-/**
- * Bearer wins over SigV4 and explicit static credentials win over the default
- * chain, matching the AWS SDK's own precedence for `AWS_BEARER_TOKEN_BEDROCK`.
- * The region is applied to the SigV4 scope so it always matches the endpoint host.
- */
-export const resolveAuth = (input: ResolveAuthInput, region: string, options: ResolveAuthOptions = {}) => {
-  const apiKey = options.mode === "sigv4" ? undefined : (input.apiKey ?? process.env.AWS_BEARER_TOKEN_BEDROCK)
-  if (apiKey !== undefined) return Auth.bearer(apiKey)
-  if (input.credentials !== undefined) return sigV4({ ...input.credentials, region }, options)
-  return sigV4(defaultChain({ region, profile: input.profile }), options)
-}
-
 export * as BedrockAuth from "./bedrock-auth.js"

packages/ai/src/protocols/utils/bedrock-cache.ts

@@ -1,6 +1,6 @@
 import { Schema } from "effect"
 import type { CacheHint } from "../../schema/index.js"
-import { newBreakpoints, ttlBucket } from "./cache.js"
+import { newBreakpoints, ttlBucket, type Breakpoints } from "./cache.js"
 
 // Bedrock cache markers are positional: emit a `cachePoint` block immediately
 // after the content the caller wants treated as a cacheable prefix. Bedrock
@@ -13,46 +13,24 @@ export const CachePointBlock = Schema.Struct({
 })
 export type CachePointBlock = Schema.Schema.Type<typeof CachePointBlock>
 
-const LEGACY_CLAUDE = ["anthropic.claude-instant", "anthropic.claude-v1", "anthropic.claude-v2", "anthropic.claude-3-"]
-
-// These legacy Claude releases support explicit caching, but only for five minutes.
-const CLAUDE_5M = [
-  "anthropic.claude-3-5-sonnet-20241022-v2:0",
-  "anthropic.claude-3-5-haiku-20241022-v1:0",
-  "anthropic.claude-3-7-sonnet-20250219-v1:0",
-  "anthropic.claude-sonnet-4-20250514-v1:0",
-  "anthropic.claude-opus-4-20250514-v1:0",
-  "anthropic.claude-opus-4-1-20250805-v1:0",
-]
-
-// Callers share the four-breakpoint budget across system, messages, and tools.
+// Callers pass a shared counter through every `block()` call site so the
+// four-breakpoint budget is respected across `system`, `messages`, and `tools`.
 export const BEDROCK_BREAKPOINT_CAP = 4
 
-export const breakpoints = (modelID: string) => {
-  // Substring matching also handles regional prefixes and model-bearing ARNs.
-  const short = CLAUDE_5M.some((id) => modelID.includes(id))
-  return {
-    ...newBreakpoints(BEDROCK_BREAKPOINT_CAP),
-    // Assume modern Claude releases retain caching support; older generations need an explicit exception.
-    // Other model families use implicit caching where available.
-    supported: modelID.includes("anthropic.claude-") && (short || !LEGACY_CLAUDE.some((id) => modelID.includes(id))),
-    ttl1h: !short,
-  }
-}
-export type Breakpoints = ReturnType<typeof breakpoints>
+export type { Breakpoints } from "./cache.js"
+export const breakpoints = () => newBreakpoints(BEDROCK_BREAKPOINT_CAP)
 
 const DEFAULT_5M: CachePointBlock = { cachePoint: { type: "default" } }
 const DEFAULT_1H: CachePointBlock = { cachePoint: { type: "default", ttl: "1h" } }
 
 export const block = (breakpoints: Breakpoints, cache: CacheHint | undefined): CachePointBlock | undefined => {
-  if (!breakpoints.supported) return undefined
   if (cache?.type !== "ephemeral" && cache?.type !== "persistent") return undefined
   if (breakpoints.remaining <= 0) {
     breakpoints.dropped += 1
     return undefined
   }
   breakpoints.remaining -= 1
-  return breakpoints.ttl1h && ttlBucket(cache.ttlSeconds) === "1h" ? DEFAULT_1H : DEFAULT_5M
+  return ttlBucket(cache.ttlSeconds) === "1h" ? DEFAULT_1H : DEFAULT_5M
 }
 
 export * as BedrockCache from "./bedrock-cache.js"

packages/ai/src/protocols/utils/bedrock-media.ts

@@ -1,4 +1,4 @@
-import { Effect, Encoding, Schema } from "effect"
+import { Effect, Schema } from "effect"
 import type { MediaPart } from "../../schema/index.js"
 import { ProviderShared } from "../shared.js"
 
@@ -57,16 +57,6 @@ const documentBlock = (name: string, format: DocumentFormat, bytes: string): Doc
   },
 })
 
-const mediaBase64 = Effect.fn("BedrockMedia.mediaBase64")(function* (part: MediaPart) {
-  const media = ProviderShared.normalizeMedia(part)
-  const bytes = yield* Effect.fromResult(Encoding.decodeBase64(media.base64)).pipe(
-    Effect.mapError((cause) =>
-      ProviderShared.invalidRequest("Bedrock Converse media data must be valid base64", cause),
-    ),
-  )
-  return Encoding.encodeBase64(bytes)
-})
-
 // Route by MIME. Known image/document formats lower into a typed block; anything
 // else fails with a clear error instead of silently degrading to a malformed
 // document block. Image MIME types not in `IMAGE_FORMATS` (e.g. `image/svg+xml`)
@@ -76,7 +66,8 @@ export const lower = Effect.fn("BedrockMedia.lower")(function* (part: MediaPart)
   const mime = part.mediaType.toLowerCase()
   const imageFormat = IMAGE_FORMATS[mime as keyof typeof IMAGE_FORMATS]
   if (imageFormat) {
-    return { image: { format: imageFormat, source: { bytes: yield* mediaBase64(part) } } } satisfies ImageBlock
+    const media = ProviderShared.normalizeMedia(part)
+    return { image: { format: imageFormat, source: { bytes: media.base64 } } } satisfies ImageBlock
   }
   if (mime.startsWith("image/"))
     return yield* ProviderShared.invalidRequest(`Bedrock Converse does not support image media type ${part.mediaType}`)
@@ -84,7 +75,8 @@ export const lower = Effect.fn("BedrockMedia.lower")(function* (part: MediaPart)
   if (documentFormat) {
     if (!part.filename)
       return yield* ProviderShared.invalidRequest("Bedrock Converse document media requires a filename")
-    return documentBlock(part.filename, documentFormat, yield* mediaBase64(part))
+    const media = ProviderShared.normalizeMedia(part)
+    return documentBlock(part.filename, documentFormat, media.base64)
   }
   return yield* ProviderShared.invalidRequest(`Bedrock Converse does not support media type ${part.mediaType}`)
 })

packages/ai/src/protocols/utils/lifecycle.ts

@@ -62,31 +62,22 @@ export const reasoningEnd = (
   events: LLMEvent[],
   id: string,
   providerMetadata?: ProviderMetadata,
-  /** Authoritative complete value; replaces accumulated deltas when present. */
-  text?: string,
 ): State => {
   if (!state.reasoning.has(id)) return state
   const stepped = stepStart(state, events)
-  events.push(LLMEvent.reasoningEnd({ id, text, providerMetadata }))
+  events.push(LLMEvent.reasoningEnd({ id, providerMetadata }))
   const reasoning = new Set(stepped.reasoning)
   reasoning.delete(id)
   return { ...stepped, reasoning }
 }
 
-export const textEnd = (
-  state: State,
-  events: LLMEvent[],
-  id: string,
-  providerMetadata?: ProviderMetadata,
-  /** Authoritative complete value; replaces accumulated deltas when present. */
-  text?: string,
-): State => {
+export const textEnd = (state: State, events: LLMEvent[], id: string, providerMetadata?: ProviderMetadata): State => {
   if (!state.text.has(id)) return state
   const stepped = stepStart(state, events)
-  events.push(LLMEvent.textEnd({ id, text, providerMetadata }))
-  const open = new Set(stepped.text)
-  open.delete(id)
-  return { ...stepped, text: open }
+  events.push(LLMEvent.textEnd({ id, providerMetadata }))
+  const text = new Set(stepped.text)
+  text.delete(id)
+  return { ...stepped, text }
 }
 
 const closeOpenBlocks = (state: State, events: LLMEvent[]): State => {

packages/ai/src/protocols/utils/meta-image.ts

@@ -1,11 +0,0 @@
-// Responses image items can omit output_format, including when PNG/JPEG was requested.
-export const mediaType = (data: Uint8Array, format?: string) => {
-  if (format !== undefined) return `image/${format}`
-  if (data[0] === 137 && data[1] === 80 && data[2] === 78 && data[3] === 71) return "image/png"
-  if (data[0] === 255 && data[1] === 216 && data[2] === 255) return "image/jpeg"
-  if (new TextDecoder().decode(data.slice(0, 4)) === "RIFF" && new TextDecoder().decode(data.slice(8, 12)) === "WEBP")
-    return "image/webp"
-  return "application/octet-stream"
-}
-
-export * as MetaImage from "./meta-image.js"

packages/ai/src/protocols/utils/responses-checkpoint.ts

@@ -1,122 +0,0 @@
-import { Effect, Schema, Stream } from "effect"
-import { Route, type RouteBody, type TriggerCompactOperation } from "../../route/client.js"
-import { Protocol } from "../../route/protocol.js"
-import { CompactionCheckpointResponse, HttpOptions, LLMEvent, LLMRequest } from "../../schema/index.js"
-import { OpenResponses } from "../open-responses.js"
-import { ProviderShared } from "../shared.js"
-
-interface State {
-  readonly parser: Pick<OpenResponses.ParserState, "id" | "provider" | "outputItems">
-  readonly checkpoints: Readonly<Record<string, CompactionCheckpointResponse["checkpoint"]>>
-  readonly responseID?: string
-}
-
-const onOutputItem = Effect.fn("ResponsesCheckpoint.onOutputItem")(function* (
-  state: State,
-  input: OpenResponses.Event,
-) {
-  const event = OpenResponses.normalize(state.parser, input)
-  const item = event.item
-  if (!item) return state
-  const parser =
-    event.output_index === undefined || state.parser.outputItems[event.output_index] === item.id
-      ? state.parser
-      : { ...state.parser, outputItems: { ...state.parser.outputItems, [event.output_index]: item.id } }
-  const next = parser === state.parser ? state : { ...state, parser }
-  if (event.type === "response.output_item.added" || item.type !== "compaction") return next
-  if (
-    event.output_index !== undefined &&
-    Object.entries(state.parser.outputItems).some(
-      ([index, id]) => id === item.id && Number(index) !== event.output_index,
-    )
-  )
-    return yield* ProviderShared.eventError(parser.id, "Compaction checkpoint appeared in multiple output slots")
-  if (!item.encrypted_content)
-    return yield* ProviderShared.eventError(parser.id, "Compaction output is missing its encrypted content")
-  const previous = state.checkpoints[item.id]
-  if (previous && previous.encrypted !== item.encrypted_content)
-    return yield* ProviderShared.eventError(parser.id, "Compaction output changed after completion")
-  if (previous) return next
-  return {
-    ...next,
-    checkpoints: {
-      ...state.checkpoints,
-      [item.id]: { type: "compaction", provider: parser.provider, id: item.id, encrypted: item.encrypted_content },
-    },
-  } satisfies State
-})
-
-/** Collect a trigger response before acknowledging transport completion. No generation output escapes. */
-export const make = <Body>(body: RouteBody<Body>): TriggerCompactOperation =>
-  Effect.fn("ResponsesCheckpoint.execute")(function* (request, executor, options) {
-    const source = request.model.route
-    let result: CompactionCheckpointResponse | undefined
-    // Route registries erase the frame type. The codec validates that boundary before parsing.
-    const event: Schema.Codec<OpenResponses.Event, unknown> = OpenResponses.protocol.stream.event
-    const protocol = Protocol.make({
-      id: source.protocol,
-      body,
-      stream: {
-        event,
-        initial: (request: LLMRequest): State => ({
-          parser: { id: source.id, provider: request.model.provider, outputItems: {} },
-          checkpoints: {},
-        }),
-        terminal: OpenResponses.terminal,
-        step: Effect.fn("ResponsesCheckpoint.step")(function* (state: State, event: OpenResponses.Event) {
-          if (event.response?.id && state.responseID && event.response.id !== state.responseID)
-            return yield* ProviderShared.eventError(source.id, "Compaction response ID changed during execution")
-          if (event.type === "response.created") return [{ ...state, responseID: event.response?.id }, []] as const
-          if (event.type === "error" || event.type === "response.failed")
-            return yield* OpenResponses.providerFailure(event, "Compaction request failed")
-          if (event.type === "response.incomplete")
-            return yield* ProviderShared.eventError(source.id, "Compaction response was incomplete")
-          if (event.type === "response.output_item.added" || event.type === "response.output_item.done")
-            return [yield* onOutputItem(state, event), []] as const
-          if (event.type !== "response.completed") return [state, []] as const
-          const responseID = event.response?.id
-          if (!responseID?.trim())
-            return yield* ProviderShared.eventError(source.id, "Compaction response is missing its response ID")
-          if (event.response?.status !== undefined && event.response.status !== "completed")
-            return yield* ProviderShared.eventError(source.id, "Compaction response did not complete successfully")
-          let next = state
-          for (const [index, item] of (event.response?.output ?? []).entries()) {
-            next = yield* onOutputItem(next, { type: "response.output_item.done", output_index: index, item })
-          }
-          const checkpoints = Object.values(next.checkpoints)
-          const checkpoint = checkpoints[0]
-          if (checkpoints.length !== 1 || !checkpoint)
-            return yield* ProviderShared.eventError(
-              source.id,
-              "Compaction response must contain exactly one checkpoint",
-            )
-          result = new CompactionCheckpointResponse({
-            checkpoint,
-            responseID,
-            usage: OpenResponses.mapUsage(event.response?.usage, OpenResponses.metadataKey(request.model)),
-          })
-          return [next, [LLMEvent.finish({ reason: { normalized: "stop" } })]] as const
-        }),
-      },
-    })
-    const route = Route.make({
-      id: source.id,
-      provider: source.provider,
-      providerMetadataKey: source.providerMetadataKey,
-      protocol,
-      endpoint: source.endpoint,
-      auth: source.auth,
-      transport: source.transport,
-    })
-    const native = yield* body.from(request)
-    // The body builder already applied and validated overlays. Do not let transport reapply them.
-    const preparedRequest = LLMRequest.update(request, {
-      http: request.http === undefined ? undefined : new HttpOptions({ ...request.http, body: undefined }),
-    })
-    const prepared = yield* route.prepareTransport(native, preparedRequest, options)
-    yield* route.streamPrepared(prepared, preparedRequest, { http: executor }, options).pipe(Stream.runDrain)
-    if (!result) return yield* ProviderShared.eventError(source.id, "Compaction response ended without a checkpoint")
-    return result
-  })
-
-export * as ResponsesCheckpoint from "./responses-checkpoint.js"

packages/ai/src/protocols/utils/responses-compaction.ts

@@ -1,174 +0,0 @@
-import { Effect, Schema, Stream } from "effect"
-import {
-  AIError,
-  InvalidProviderOutputError,
-  CompactionPart,
-  CompactionResponse,
-  HttpOptions,
-  LLMRequest,
-  Message,
-  type ContentPart,
-  mergeJsonRecords,
-} from "../../schema/index.js"
-import type { CompactOperation } from "../../route/client.js"
-import { Endpoint } from "../../route/endpoint.js"
-import { RequestExecutor } from "../../route/executor.js"
-import { HttpTransport } from "../../route/transport/index.js"
-import { OpenResponses } from "../open-responses.js"
-import { JsonObject, optionalNull, ProviderShared } from "../shared.js"
-
-const Body = Schema.Struct({
-  model: Schema.String,
-  input: Schema.Array(Schema.Unknown),
-  instructions: optionalNull(Schema.String),
-  previous_response_id: optionalNull(Schema.String),
-  service_tier: optionalNull(Schema.String),
-  prompt_cache_key: optionalNull(Schema.String),
-  prompt_cache_retention: optionalNull(Schema.String),
-  prompt_cache_options: optionalNull(
-    Schema.Struct({ mode: Schema.optional(Schema.String), ttl: Schema.optional(Schema.String) }),
-  ),
-})
-
-const Text = Schema.Union([OpenResponses.OpenResponsesInputText, OpenResponses.OpenResponsesOutputText])
-const File = Schema.Union([
-  Schema.Struct({
-    ...OpenResponses.OpenResponsesInputFile.fields,
-    file_url: Schema.String,
-    file_data: Schema.optional(Schema.Never),
-  }),
-  Schema.Struct({
-    ...OpenResponses.OpenResponsesInputFile.fields,
-    file_data: Schema.String,
-    file_url: Schema.optional(Schema.Never),
-  }),
-])
-const MessageFields = {
-  type: Schema.Literal("message"),
-  id: Schema.optional(Schema.String),
-  status: Schema.optional(Schema.String),
-  phase: Schema.optional(OpenResponses.MessagePhase),
-}
-const Response = Schema.Struct({
-  object: Schema.Literal("response.compaction"),
-  output: Schema.Array(
-    Schema.Union([
-      OpenResponses.CompactionItem,
-      OpenResponses.OpenResponsesReasoningItem,
-      Schema.Struct({
-        ...MessageFields,
-        role: Schema.Literal("user"),
-        content: Schema.Array(Schema.Union([Text, OpenResponses.OpenResponsesInputImage, File])).check(
-          Schema.isMinLength(1),
-        ),
-      }),
-      Schema.Struct({
-        ...MessageFields,
-        role: Schema.Literal("assistant"),
-        content: Schema.Array(Text).check(Schema.isMinLength(1)),
-      }),
-    ]),
-  ),
-  usage: Schema.optional(Schema.StructWithRest(OpenResponses.OpenResponsesUsage, [JsonObject])),
-})
-
-export const make = (adapter: OpenResponses.ProviderAdapter): CompactOperation =>
-  Effect.fn("ResponsesCompaction.execute")(function* (request, executor, options) {
-    const route = request.model.route
-    const native = yield* OpenResponses.lowerConversation(request, adapter)
-    const body = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(Body))(
-      mergeJsonRecords(
-        {
-          ...native,
-          service_tier: request.providerOptions?.serviceTier,
-          prompt_cache_key: ProviderShared.promptCacheKey(request),
-        },
-        request.http?.body,
-      ),
-    )
-    const url = Endpoint.render(route.endpoint, { request, body: native })
-    url.pathname = `${url.pathname.replace(/\/$/, "")}/compact`
-    const parts = yield* HttpTransport.jsonRequestParts({
-      request: LLMRequest.update(request, {
-        http: request.http === undefined ? undefined : new HttpOptions({ ...request.http, body: undefined }),
-      }),
-      body,
-      endpoint: Endpoint.path(url.toString()),
-      auth: route.auth,
-      encodeBody: Schema.encodeSync(Schema.fromJsonString(Body)),
-    })
-    const response = yield* executor.execute(
-      ProviderShared.jsonPost({ url: parts.url, body: parts.bodyText, headers: parts.headers }),
-      options?.http,
-    )
-    const text = yield* RequestExecutor.responseStream(response).pipe(
-      Stream.decodeText(),
-      Stream.runFold(
-        () => "",
-        (text, chunk) => text + chunk,
-      ),
-    )
-    const invalid = (message: string, cause?: unknown) =>
-      new AIError({
-        reason: new InvalidProviderOutputError({
-          route: route.id,
-          message,
-          body: text,
-          cause,
-          http: RequestExecutor.responseHttp(response),
-        }),
-      })
-    const result = yield* Schema.decodeUnknownEffect(Schema.fromJsonString(Response))(text).pipe(
-      Effect.mapError((cause) => invalid("Invalid compaction response", cause)),
-    )
-    if (!result.output.some((item) => item.type === "compaction"))
-      return yield* invalid("Compaction response did not contain a checkpoint")
-    return new CompactionResponse({
-      replacement: result.output.map((item) => toMessage(item, request.model)),
-      usage: OpenResponses.mapUsage(result.usage, OpenResponses.metadataKey(request.model)),
-    })
-  })
-
-function toMessage(item: (typeof Response.Type.output)[number], model: LLMRequest["model"]): Message {
-  if (item.type === "compaction")
-    return Message.assistant(
-      CompactionPart.make({ provider: model.provider, id: item.id ?? undefined, encrypted: item.encrypted_content }),
-    )
-
-  const key = OpenResponses.metadataKey(model)
-  if (item.type === "reasoning") {
-    const summary = item.summary.length ? item.summary : [{ text: "" }]
-    return Message.assistant(
-      summary.map((part) => ({
-        type: "reasoning" as const,
-        text: part.text,
-        providerMetadata: { [key]: { itemId: item.id, reasoningEncryptedContent: item.encrypted_content } },
-      })),
-    )
-  }
-
-  return Message.make({
-    role: item.role,
-    providerMetadata: { [key]: { itemId: item.id, type: item.type, status: item.status, phase: item.phase } },
-    content: item.content.map((part): ContentPart => {
-      if (part.type === "input_text" || part.type === "output_text") return { type: "text", text: part.text }
-      if (part.type === "input_image")
-        return {
-          type: "media",
-          data: part.image_url,
-          mediaType: /^data:([^;,]+)/.exec(part.image_url)?.[1] ?? "image/*",
-          providerMetadata: part.detail === undefined ? undefined : { [key]: { detail: part.detail } },
-        }
-      const data = part.file_url === undefined ? part.file_data : part.file_url
-      return {
-        type: "media",
-        data,
-        filename: part.filename,
-        mediaType: /^data:([^;,]+)/.exec(data)?.[1] ?? "application/octet-stream",
-        providerMetadata: part.detail === undefined ? undefined : { [key]: { detail: part.detail } },
-      }
-    }),
-  })
-}
-
-export * as ResponsesCompaction from "./responses-compaction.js"

packages/ai/src/protocols/utils/responses-hosted-tools.ts

@@ -3,7 +3,8 @@ import { LLMEvent, type AIError, type ToolResultPart } from "../../schema/index.
 import { OpenResponses } from "../open-responses.js"
 import { Lifecycle } from "./lifecycle.js"
 
-export type Item = OpenResponses.OutputItem & {
+export type Item = OpenResponses.StreamItem & {
+  readonly id: string
   readonly status?: string
   readonly action?: unknown
   readonly queries?: unknown
@@ -26,8 +27,8 @@ export interface Definition {
 
 export type Definitions = Readonly<Record<string, Definition>>
 
-export const isItem = <Tools extends Definitions>(item: OpenResponses.OutputItem, tools: Tools): item is Item =>
-  item.type in tools
+export const isItem = <Tools extends Definitions>(item: OpenResponses.StreamItem, tools: Tools): item is Item =>
+  item.type in tools && typeof item.id === "string" && item.id.length > 0
 
 export const onDone: (
   state: OpenResponses.ParserState,

packages/ai/src/protocols/utils/tool-stream.ts

@@ -55,7 +55,6 @@ const inputStart = (tool: PendingTool) =>
   LLMEvent.toolInputStart({
     id: tool.id,
     name: tool.name,
-    namespace: tool.namespace,
     providerExecuted: tool.providerExecuted ? true : undefined,
     providerMetadata: tool.providerMetadata,
   })
@@ -64,7 +63,6 @@ const inputDelta = (tool: PendingTool, text: string) =>
   LLMEvent.toolInputDelta({
     id: tool.id,
     name: tool.name,
-    namespace: tool.namespace,
     text,
     input: Option.getOrElse(parsePartialInput(tool.input), () => ({})),
   })
@@ -87,7 +85,6 @@ const toolCall = (route: string, tool: PendingTool, inputOverride?: string) => {
         LLMEvent.toolCall({
           id: tool.id,
           name: tool.name,
-          namespace: tool.namespace,
           input,
           providerExecuted: tool.providerExecuted ? true : undefined,
           providerMetadata: tool.providerMetadata,
@@ -97,12 +94,7 @@ const toolCall = (route: string, tool: PendingTool, inputOverride?: string) => {
 }
 
 const finishEvents = (tool: PendingTool, event: ToolCall): ReadonlyArray<LLMEvent> => [
-  LLMEvent.toolInputEnd({
-    id: tool.id,
-    name: tool.name,
-    namespace: tool.namespace,
-    providerMetadata: tool.providerMetadata,
-  }),
+  LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }),
   event,
 ]
 
@@ -158,7 +150,6 @@ export const appendOrStart = <K extends StreamKey>(
   const tool = {
     id,
     name,
-    namespace: current?.namespace,
     input: `${current?.input ?? ""}${delta.text}`,
     providerExecuted: current?.providerExecuted,
     providerMetadata: current?.providerMetadata,
@@ -168,17 +159,6 @@ export const appendOrStart = <K extends StreamKey>(
   return appendTool(tools, key, tool, delta.text)
 }
 
-/**
- * Append argument text to a started tool. Returns `undefined` when no tool is
- * open under `key`, for protocols that ignore deltas without a matching block.
- */
-export const append = <K extends StreamKey>(tools: State<K>, key: K, text: string): AppendOutcome<K> | undefined => {
-  const current = tools[key]
-  if (!current) return undefined
-  if (text.length === 0) return { tools, tool: current, events: [] }
-  return appendTool(tools, key, { ...current, input: `${current.input}${text}` }, text)
-}
-
 /**
  * Append argument text to a tool that must already have been started. This keeps
  * protocols honest when their stream grammar promises a start event before any
@@ -190,7 +170,12 @@ export const appendExisting = <K extends StreamKey>(
   key: K,
   text: string,
   missingToolMessage: string,
-): AppendOutcome<K> | AIError => append(tools, key, text) ?? eventError(route, missingToolMessage)
+): AppendOutcome<K> | AIError => {
+  const current = tools[key]
+  if (!current) return eventError(route, missingToolMessage)
+  if (text.length === 0) return { tools, tool: current, events: [] }
+  return appendTool(tools, key, { ...current, input: `${current.input}${text}` }, text)
+}
 
 /**
  * Finalize one pending tool call: parse the accumulated raw JSON, remove it

packages/ai/src/protocols/xai-responses.ts

@@ -4,7 +4,6 @@ import type { LLMRequest } from "../schema/index.js"
 import { OpenResponses } from "./open-responses.js"
 import { JsonObject, optionalNull, ProviderShared } from "./shared.js"
 import { ResponsesHostedTools } from "./utils/responses-hosted-tools.js"
-import { ResponsesCompaction } from "./utils/responses-compaction.js"
 
 const ADAPTER = "xai-responses"
 const NAME = "xAI Responses"
@@ -37,22 +36,15 @@ const XAIResponsesBody = Schema.Struct({
   stream: Schema.Literal(true),
 })
 
-const adapter = {
+const extension = {
   id: ADAPTER,
   name: NAME,
-  restoreHostedToolItem: (item: unknown) => (Schema.is(XAIResponsesHostedToolItem)(item) ? item : undefined),
-} satisfies OpenResponses.ProviderAdapter
+  lowerHostedToolItem: (item: unknown) => (Schema.is(XAIResponsesHostedToolItem)(item) ? item : undefined),
+} satisfies OpenResponses.Extension
 
 const decodeBody = ProviderShared.validateWith(Schema.decodeUnknownEffect(XAIResponsesBody))
 const fromRequest = Effect.fn("XAIResponses.fromRequest")(function* (request: LLMRequest) {
-  if (request.providerOptions?.contextManagement !== undefined)
-    return yield* ProviderShared.unsupportedOperation({
-      operation: "in-band-compaction",
-      provider: request.model.provider,
-      route: request.model.route.id,
-      message: "xAI requires explicit compaction through LLMClient.compact; automatic context management is not supported",
-    })
-  return yield* decodeBody(yield* OpenResponses.fromRequestWithAdapter(request, adapter))
+  return yield* decodeBody(yield* OpenResponses.fromRequestWithExtension(request, extension))
 })
 
 const HOSTED_TOOLS = {
@@ -72,8 +64,7 @@ const HOSTED_TOOLS = {
 
 // Grok speaks the standard Responses reasoning dialect (`reasoning_summary_text.*`,
 // handled by the baseline); only its hosted tool vocabulary differs.
-const step = (state: OpenResponses.ParserState, input: OpenResponses.Event) => {
-  const event = OpenResponses.normalize(state, input)
+const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {
   if (event.type === "response.output_item.done" && event.item && ResponsesHostedTools.isItem(event.item, HOSTED_TOOLS))
     return ResponsesHostedTools.onDone(state, event.item, HOSTED_TOOLS)
   return OpenResponses.step(state, event)
@@ -87,12 +78,10 @@ export const protocol = Protocol.make({
   },
   stream: {
     event: OpenResponses.protocol.stream.event,
-    initial: (request) => OpenResponses.initial(request, adapter),
+    initial: (request) => OpenResponses.initial(request, extension),
     step,
     terminal: OpenResponses.terminal,
   },
 })
 
-export const compact = ResponsesCompaction.make(adapter)
-
 export * as XAIResponses from "./xai-responses.js"

packages/ai/src/provider-error.ts

@@ -37,7 +37,6 @@ const patterns = [
   /too large for model with \d+ maximum context length/i,
   /prompt has [\d,]+ tokens?, but the configured context size is [\d,]+ tokens?/i,
   /model_context_window_exceeded/i,
-  /range of input length should be/i,
   /too many tokens/i,
   /token limit exceeded/i,
   /request_too_large/i,
@@ -60,7 +59,6 @@ export const isContextOverflowFailure = (failure: unknown) =>
 
 const decodeJson = Schema.decodeUnknownOption(Schema.fromJsonString(Schema.Unknown))
 const QUOTA_CODES = new Set(["insufficient_quota", "usage_not_included", "billing_error"])
-const AUTH_CODES = new Set(["authentication_error", "permission_error"])
 const SERVER_CODES = new Set([
   "api_error",
   "internal_error",
@@ -76,8 +74,7 @@ const INVALID_REQUEST_CODES = new Set(["invalid_prompt", "invalid_request_error"
 const RATE_LIMIT_TEXT = /rate increased too quickly|rate[-_\s]?limit|too[_\s]?many[_\s]?requests/i
 const QUOTA_TEXT = /insufficient[-_\s]?quota|quota[-_\s]?exceeded/i
 const CONTENT_POLICY_TEXT = /content[-_\s]?policy|content_filter|safety/i
-const SERVER_ERROR_TEXT =
-  /\b(?:try again|(?:please |you can )?retry (?:the |this |your )?request|try (?:the |this |your )?request again|(?:currently |temporarily )?at capacity|overloaded|temporarily unavailable|service[-_\s]?unavailable|(?:server|internal)[-_\s]?error|server (?:is )?busy|provider returned (?:an )?error|resource[-_\s]?exhausted|upstream (?:connect|connection|request)|request buffer limit while retrying upstream)\b/i
+const NETWORK_ERROR_TEXT = /network[-_\s]error/i
 
 export interface ProviderFailure {
   readonly message: string
@@ -93,12 +90,8 @@ export interface ProviderFailure {
   readonly rateLimit?: HttpRateLimitDetails | undefined
 }
 
-// Classification records affirmative evidence about a failure. Deterministic
-// failures need positive identification (a 4xx status, quota/auth/policy
-// signals); anything unrecognized stays UnknownProvider, which the session
-// retry policy treats as retry-eligible because transient failures arrive in
-// unpredictable shapes while deterministic rejections almost always carry a
-// status or known code.
+// Keep HTTP failures and provider-reported stream failures on one typed path so
+// session retry policy never needs provider-specific string matching.
 export function classifyProviderFailure(input: ProviderFailure): AIError["reason"] {
   const details = { message: input.message, body: input.rawBody, http: input.http, cause: input.cause }
   const body = input.rawBody ?? ""
@@ -123,36 +116,46 @@ export function classifyProviderFailure(input: ProviderFailure): AIError["reason
   if (CONTENT_POLICY_TEXT.test(text)) return new ContentPolicyError(details)
   if (codes.some((code) => QUOTA_CODES.has(code)) || (input.status === 429 && QUOTA_TEXT.test(text)))
     return new QuotaExceededError(details)
-  if (input.status === 401 || input.status === 403 || codes.some((code) => AUTH_CODES.has(code)))
-    return new AuthenticationError(details)
+  if (input.status === 401) return new AuthenticationError({ ...details, kind: "invalid" })
+  if (input.status === 403) return new AuthenticationError({ ...details, kind: "insufficient-permissions" })
+  if (codes.includes("authentication_error")) return new AuthenticationError({ ...details, kind: "invalid" })
+  if (codes.includes("permission_error"))
+    return new AuthenticationError({ ...details, kind: "insufficient-permissions" })
   if (
-    input.status === 429 ||
-    codes.some(
-      (code) => code.includes("rate_limit") || code === "too_many_requests" || code === "throttlingexception",
-    ) ||
-    RATE_LIMIT_TEXT.test(text)
+    codes.some((code) => code.includes("rate_limit") || code === "too_many_requests" || code === "throttlingexception")
   )
     return new RateLimitError({
       ...details,
       retryAfterMs: input.retryAfterMs,
       rateLimit: input.rateLimit,
     })
-  if (
-    input.status === 408 ||
-    input.status === 409 ||
-    (input.status !== undefined && input.status >= 500) ||
-    ((input.status === undefined || input.status < 400) &&
-      !codes.some((code) => INVALID_REQUEST_CODES.has(code)) &&
-      SERVER_ERROR_TEXT.test(text)) ||
-    codes.some((code) => SERVER_CODES.has(code) || code.includes("exhausted") || code.includes("unavailable"))
-  )
+  if (RATE_LIMIT_TEXT.test(text))
+    return new RateLimitError({
+      ...details,
+      retryAfterMs: input.retryAfterMs,
+      rateLimit: input.rateLimit,
+    })
+  if (NETWORK_ERROR_TEXT.test(text)) return new ProviderInternalError(details)
+  if (codes.some((code) => SERVER_CODES.has(code) || code.includes("exhausted") || code.includes("unavailable")))
+    return new ProviderInternalError({
+      ...details,
+      retryAfterMs: input.retryAfterMs,
+    })
+  if (input.status === 429) {
+    return new RateLimitError({
+      ...details,
+      retryAfterMs: input.retryAfterMs,
+      rateLimit: input.rateLimit,
+    })
+  }
+  if (input.status === 408 || input.status === 409 || (input.status !== undefined && input.status >= 500))
     return new ProviderInternalError({
       ...details,
       retryAfterMs: input.retryAfterMs,
     })
   if (codes.some((code) => INVALID_REQUEST_CODES.has(code))) return new InvalidRequestError(details)
-  // Any remaining 4xx is a deterministic rejection of this request.
-  if (input.status !== undefined && input.status >= 400 && input.status < 500) return new InvalidRequestError(details)
+  if (input.status === 400 || input.status === 404 || input.status === 413 || input.status === 422)
+    return new InvalidRequestError(details)
   return new UnknownProviderError(details)
 }
 

packages/ai/src/provider-package.ts

@@ -1,5 +1,4 @@
 import type { LanguageModel, ProviderOptions } from "./schema/index.js"
-import type { CompactionOperations } from "./route/client.js"
 
 export interface Settings extends Readonly<Record<string, unknown>> {
   readonly baseURL?: string
@@ -10,9 +9,8 @@ export interface Settings extends Readonly<Record<string, unknown>> {
 export interface Definition<
   ProviderSettings extends Settings = Settings,
   Options extends ProviderOptions = ProviderOptions,
-  Compact extends CompactionOperations | undefined = CompactionOperations | undefined,
 > {
-  readonly model: (modelID: string, settings: ProviderSettings) => LanguageModel<Options, Compact>
+  readonly model: (modelID: string, settings: ProviderSettings) => LanguageModel<Options>
 }
 
 export * as ProviderPackage from "./provider-package.js"

packages/ai/src/providers/amazon-bedrock-mantle.ts

@@ -1,3 +1,4 @@
+import { Auth } from "../route/auth.js"
 import { Route, type RouteDefaultsInput } from "../route/client.js"
 import type { ProviderPackage } from "../provider-package.js"
 import { OpenAIChat } from "../protocols/openai-chat.js"
@@ -9,15 +10,9 @@ import { withOpenAIOptions, type OpenAIProviderOptionsInput } from "./openai-opt
 export const id = ProviderID.make("amazon-bedrock")
 
 export type Config = RouteDefaultsInput & {
-  /** Bedrock API key. Falls back to `AWS_BEARER_TOKEN_BEDROCK`; bearer auth takes precedence over SigV4. */
   readonly apiKey?: string
-  /** `sigv4` ignores `apiKey` fallbacks from the environment; `bearer` requires a token. */
-  readonly auth?: "bearer" | "sigv4"
   readonly baseURL?: string
-  /** Static SigV4 credentials. When omitted the AWS default credential chain resolves them per request. */
   readonly credentials?: Credentials
-  /** Shared config profile for the default credential chain. */
-  readonly profile?: string
   readonly region?: string
   readonly providerOptions?: OpenAIProviderOptionsInput
 }
@@ -27,7 +22,6 @@ export interface Settings extends ProviderPackage.Settings {
   readonly auth?: "bearer" | "sigv4"
   readonly baseURL?: string
   readonly credentials?: Credentials
-  readonly profile?: string
   readonly region?: string
   readonly topP?: number
   readonly providerOptions?: OpenAIProviderOptionsInput
@@ -53,35 +47,23 @@ const chatRoute = OpenAIChat.route.with({
 export const routes = [responsesRoute, chatRoute]
 
 const configuredRoute = <Body, Prepared>(route: Route<Body, Prepared>, input: Config) => {
-  const region = BedrockAuth.resolveRegion(input)
+  const region = input.region ?? input.credentials?.region ?? "us-east-1"
+  const credentials = input.credentials === undefined ? undefined : { ...input.credentials, region }
   return route.with({
     endpoint: { baseURL: input.baseURL ?? `https://bedrock-mantle.${region}.api.aws/v1` },
-    auth: BedrockAuth.resolveAuth(input, region, {
-      service: "bedrock-mantle",
-      name: "Bedrock Mantle",
-      mode: input.auth,
-    }),
+    auth:
+      input.apiKey === undefined
+        ? BedrockAuth.sigV4(credentials, { service: "bedrock-mantle", name: "Bedrock Mantle" })
+        : Auth.bearer(input.apiKey),
   })
 }
 
 const defaults = (input: Config) => {
-  const {
-    apiKey: _,
-    auth: _auth,
-    baseURL: _baseURL,
-    credentials: _credentials,
-    profile: _profile,
-    region: _region,
-    ...rest
-  } = input
+  const { apiKey: _, baseURL: _baseURL, credentials: _credentials, region: _region, ...rest } = input
   return rest
 }
 
 export const configure = (input: Config = {}) => {
-  if (input.auth === "bearer" && input.apiKey === undefined && process.env.AWS_BEARER_TOKEN_BEDROCK === undefined)
-    throw new Error("Amazon Bedrock Mantle bearer auth requires apiKey")
-  if (input.auth === "sigv4" && input.apiKey !== undefined)
-    throw new Error("Amazon Bedrock Mantle SigV4 auth does not accept apiKey")
   const configuredResponsesRoute = configuredRoute(responsesRoute, input)
   const configuredChatRoute = configuredRoute(chatRoute, input)
   const modelDefaults = defaults(input)
@@ -105,26 +87,29 @@ export const configure = (input: Config = {}) => {
 
 export const provider = configure()
 
-const fromSettings = (settings: Settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    auth: settings.auth,
+const config = (settings: Settings): Config => {
+  if (settings.auth === "bearer" && settings.apiKey === undefined)
+    throw new Error("Amazon Bedrock Mantle bearer auth requires apiKey")
+  if (settings.auth === "sigv4" && settings.apiKey !== undefined)
+    throw new Error("Amazon Bedrock Mantle SigV4 auth does not accept apiKey")
+  return {
+    apiKey: settings.auth === "sigv4" ? undefined : settings.apiKey,
     baseURL: settings.baseURL,
     credentials: settings.credentials,
     generation: settings.topP === undefined ? undefined : { topP: settings.topP },
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    profile: settings.profile,
     providerOptions: settings.providerOptions,
     region: settings.region,
-  })
+  }
+}
 
 export const chatModel: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (
   modelID,
   settings,
-) => fromSettings(settings).chat(modelID)
+) => configure(config(settings)).chat(modelID)
 export const responsesModel: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (
   modelID,
   settings,
-) => fromSettings(settings).responses(modelID)
+) => configure(config(settings)).responses(modelID)
 export const model = responsesModel

packages/ai/src/providers/amazon-bedrock.ts

@@ -1,23 +1,17 @@
 import type { RouteDefaultsInput } from "../route/client.js"
+import { Auth } from "../route/auth.js"
 import type { ProviderPackage } from "../provider-package.js"
 import { ProviderID, type ModelID } from "../schema/index.js"
 import * as BedrockConverse from "../protocols/bedrock-converse.js"
 import type { BedrockCredentials } from "../protocols/bedrock-converse.js"
-import { BedrockAuth } from "../protocols/utils/bedrock-auth.js"
 
 export const id = ProviderID.make("amazon-bedrock")
 
 export type Config = RouteDefaultsInput & {
-  /** Bedrock API key. Falls back to `AWS_BEARER_TOKEN_BEDROCK`; bearer auth takes precedence over SigV4. */
   readonly apiKey?: string
-  /** `sigv4` ignores `apiKey` fallbacks from the environment; `bearer` requires a token. */
-  readonly auth?: "bearer" | "sigv4"
   readonly headers?: Record<string, string>
-  /** Static SigV4 credentials. When omitted the AWS default credential chain resolves them per request. */
   readonly credentials?: BedrockCredentials
-  /** Shared config profile for the default credential chain. */
-  readonly profile?: string
-  /** AWS region. Falls back to `credentials.region`, `AWS_REGION`, `AWS_DEFAULT_REGION`, then `us-east-1`. */
+  /** AWS region. Defaults to `us-east-1` when neither this nor `credentials.region` is set. */
   readonly region?: string
   /** Override the computed `https://bedrock-runtime.<region>.amazonaws.com` URL. */
   readonly baseURL?: string
@@ -28,7 +22,6 @@ export interface Settings extends ProviderPackage.Settings {
   readonly auth?: "bearer" | "sigv4"
   readonly baseURL?: string
   readonly credentials?: BedrockCredentials
-  readonly profile?: string
   readonly region?: string
   readonly topP?: number
 }
@@ -37,17 +30,14 @@ export const routes = [BedrockConverse.route]
 const bedrockBaseURL = (region: string) => `https://bedrock-runtime.${region}.amazonaws.com`
 
 const configuredRoute = (input: Config) => {
-  const { apiKey, auth, credentials, profile, region, baseURL, ...rest } = input
-  if (auth === "bearer" && apiKey === undefined && process.env.AWS_BEARER_TOKEN_BEDROCK === undefined)
-    throw new Error("Amazon Bedrock bearer auth requires apiKey")
-  if (auth === "sigv4" && apiKey !== undefined) throw new Error("Amazon Bedrock SigV4 auth does not accept apiKey")
-  const resolvedRegion = BedrockAuth.resolveRegion(input)
+  const { apiKey, credentials, region, baseURL, ...rest } = input
+  const resolvedRegion = region ?? credentials?.region ?? "us-east-1"
   return BedrockConverse.route.with({
     ...rest,
     provider: id,
     providerMetadataKey: "bedrock",
     endpoint: { baseURL: baseURL ?? bedrockBaseURL(resolvedRegion) },
-    auth: BedrockAuth.resolveAuth({ apiKey, credentials, profile }, resolvedRegion, { mode: auth }),
+    auth: apiKey === undefined ? BedrockConverse.sigV4Auth(credentials) : Auth.bearer(apiKey),
   })
 }
 
@@ -61,15 +51,18 @@ export const configure = (input: Config = {}) => {
 }
 
 export const provider = configure()
-export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    auth: settings.auth,
+export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+  if (settings.auth === "bearer" && settings.apiKey === undefined)
+    throw new Error("Amazon Bedrock bearer auth requires apiKey")
+  if (settings.auth === "sigv4" && settings.apiKey !== undefined)
+    throw new Error("Amazon Bedrock SigV4 auth does not accept apiKey")
+  return configure({
+    apiKey: settings.auth === "sigv4" ? undefined : settings.apiKey,
     baseURL: settings.baseURL,
     credentials: settings.credentials,
     generation: settings.topP === undefined ? undefined : { topP: settings.topP },
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    profile: settings.profile,
     region: settings.region,
   }).model(modelID)
+}

packages/ai/src/providers/azure.ts

@@ -1,7 +1,7 @@
 import { Headers } from "effect/unstable/http"
 import { Auth } from "../route/auth.js"
 import { type AtLeastOne, type ProviderAuthOption } from "../route/auth-options.js"
-import type { Route, RouteDefaultsInput, CompactionOperations } from "../route/client.js"
+import type { Route as RouteDef, RouteDefaultsInput } from "../route/client.js"
 import type { ProviderPackage } from "../provider-package.js"
 import { ProviderID, type ModelID } from "../schema/index.js"
 import * as OpenAIChat from "../protocols/openai-chat.js"
@@ -39,7 +39,6 @@ export type Settings = ProviderPackage.Settings &
 const resourceBaseURL = (resourceName: string) => `https://${resourceName.trim()}.openai.azure.com/openai`
 
 const responsesRoute = OpenAIResponses.route.with({
-  compact: { endpoint: OpenAIResponses.route.compact.endpoint },
   id: "azure-openai-responses",
   provider: id,
   auth: routeAuth,
@@ -103,11 +102,7 @@ const auth = (input: Config) => {
   )
 }
 
-const configuredRoute = <Body, Prepared, Compact extends CompactionOperations | undefined>(
-  route: Route<Body, Prepared, Compact>,
-  input: Config,
-  modelID: string | ModelID,
-) =>
+const configuredRoute = <Body, Prepared>(route: RouteDef<Body, Prepared>, input: Config, modelID: string | ModelID) =>
   route.with({
     auth: auth(input),
     endpoint: endpoint(input, modelID),
@@ -166,11 +161,10 @@ const config = (settings: Settings): Config => {
   throw new Error("Azure requires resourceName or baseURL")
 }
 
-export const responsesModel: ProviderPackage.Definition<
-  Settings,
-  OpenAIProviderOptionsInput,
-  typeof responsesRoute.compact
->["model"] = (modelID, settings) => configure(config(settings)).responses(modelID)
+export const responsesModel: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (
+  modelID,
+  settings,
+) => configure(config(settings)).responses(modelID)
 export const chatModel: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (
   modelID,
   settings,

packages/ai/src/providers/baseten.ts

@@ -1,60 +0,0 @@
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { ProviderID, type ModelID } from "../schema/index.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("baseten")
-const baseURL = "https://inference.baseten.co/v1"
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: OpenAIProviderOptionsInput
-}
-
-export const route = Route.make({
-  id: "baseten-chat",
-  provider: id,
-  providerMetadataKey: "baseten",
-  protocol: OpenAIChat.protocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL }),
-  framing: OpenAIChat.framing,
-})
-
-export const routes = [route]
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL: endpoint, ...defaults } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: endpoint ?? baseURL },
-    auth: AuthOptions.bearer(input, "BASETEN_API_KEY"),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) => configured.model<OpenAIProviderOptionsInput>({ id: modelID }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
-
-export * as Baseten from "./baseten.js"

packages/ai/src/providers/cerebras.ts

@@ -1,13 +1,12 @@
 import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
+import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
+import type { RouteDefaultsInput } from "../route/client.js"
 import { ProviderID, type ModelID } from "../schema/index.js"
+import { profiles } from "./openai-compatible-profile.js"
 import type { OpenAIProviderOptionsInput } from "./openai-options.js"
 
 export const id = ProviderID.make("cerebras")
-const baseURL = "https://api.cerebras.ai/v1"
 
 export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
   ProviderAuthOption<"optional"> & {
@@ -21,22 +20,19 @@ export interface Settings extends ProviderPackage.Settings {
   readonly providerOptions?: OpenAIProviderOptionsInput
 }
 
-export const route = Route.make({
+export const route = OpenAICompatibleChat.route.with({
   id: "cerebras-chat",
   provider: id,
-  providerMetadataKey: "cerebras",
-  protocol: OpenAIChat.protocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL }),
-  framing: OpenAIChat.framing,
+  endpoint: { baseURL: profiles.cerebras.baseURL },
 })
 
 export const routes = [route]
 
 export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL: endpoint, ...defaults } = input
+  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
   const configured = route.with({
     ...defaults,
-    endpoint: { baseURL: endpoint ?? baseURL },
+    endpoint: { baseURL: baseURL ?? profiles.cerebras.baseURL },
     auth: AuthOptions.bearer(input, "CEREBRAS_API_KEY"),
   })
   return {

packages/ai/src/providers/cloudflare-ai-gateway.ts

@@ -1,98 +0,0 @@
-import type { Config, Redacted } from "effect"
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { Auth } from "../route/auth.js"
-import type { AtLeastOne, ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { ProviderID, type ModelID } from "../schema/index.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("cloudflare-ai-gateway")
-export const authEnvVars = ["CLOUDFLARE_API_TOKEN", "CF_AIG_TOKEN"] as const
-
-type GatewayURL = AtLeastOne<{
-  readonly accountId: string
-  readonly baseURL: string
-}> & {
-  readonly gatewayId?: string
-}
-
-export type LanguageModelOptions = GatewayURL &
-  Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    /** Cloudflare AI Gateway authentication token. Sent as `cf-aig-authorization`. */
-    readonly gatewayApiKey?: string | Redacted.Redacted | Config.Config<string | Redacted.Redacted>
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export type Settings = ProviderPackage.Settings &
-  GatewayURL & {
-    readonly apiKey?: string
-    readonly gatewayApiKey?: string
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export const baseURL = (input: GatewayURL) => {
-  if (input.baseURL) return input.baseURL
-  if (!input.accountId) throw new Error("CloudflareAIGateway.configure requires accountId unless baseURL is supplied")
-  return `https://gateway.ai.cloudflare.com/v1/${encodeURIComponent(input.accountId)}/${encodeURIComponent(input.gatewayId?.trim() || "default")}/compat`
-}
-
-const auth = (input: LanguageModelOptions) => {
-  if ("auth" in input && input.auth) return input.auth
-  const gateway = Auth.optional(input.gatewayApiKey, "gatewayApiKey")
-    .orElse(Auth.config(authEnvVars[0]))
-    .orElse(Auth.config(authEnvVars[1]))
-    .pipe(Auth.bearerHeader("cf-aig-authorization"))
-  if (!("apiKey" in input) || input.apiKey === undefined) return gateway
-  if (input.gatewayApiKey === undefined) return Auth.bearer(input.apiKey)
-  return Auth.bearerHeader("cf-aig-authorization", input.gatewayApiKey).andThen(Auth.bearer(input.apiKey))
-}
-
-export const route = Route.make({
-  id: "cloudflare-ai-gateway",
-  provider: id,
-  providerMetadataKey: "cloudflare-ai-gateway",
-  protocol: OpenAIChat.protocol,
-  endpoint: Endpoint.path("/chat/completions"),
-  framing: OpenAIChat.framing,
-})
-
-export const routes = [route]
-
-export const configure = (input: LanguageModelOptions) => {
-  const {
-    accountId: _accountId,
-    gatewayId: _gatewayId,
-    apiKey: _apiKey,
-    gatewayApiKey: _gatewayApiKey,
-    baseURL: _baseURL,
-    auth: _auth,
-    ...defaults
-  } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: baseURL(input) },
-    auth: auth(input),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) => configured.model<OpenAIProviderOptionsInput>({ id: modelID }),
-    configure,
-  }
-}
-
-export const provider = { id, configure }
-
-export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    gatewayApiKey: settings.gatewayApiKey,
-    baseURL: baseURL(settings),
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
-
-export * as CloudflareAIGateway from "./cloudflare-ai-gateway.js"

packages/ai/src/providers/cloudflare-workers-ai.ts

@@ -1,71 +0,0 @@
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { AuthOptions, type AtLeastOne, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { ProviderID, type ModelID } from "../schema/index.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("cloudflare-workers-ai")
-export const authEnvVars = ["CLOUDFLARE_API_KEY", "CLOUDFLARE_WORKERS_AI_TOKEN"] as const
-
-type WorkersAIURL = AtLeastOne<{
-  readonly accountId: string
-  readonly baseURL: string
-}>
-
-export type LanguageModelOptions = WorkersAIURL &
-  Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export type Settings = ProviderPackage.Settings &
-  WorkersAIURL & {
-    readonly apiKey?: string
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export const baseURL = (input: WorkersAIURL) => {
-  if (input.baseURL) return input.baseURL
-  if (!input.accountId) throw new Error("CloudflareWorkersAI.configure requires accountId unless baseURL is supplied")
-  return `https://api.cloudflare.com/client/v4/accounts/${encodeURIComponent(input.accountId)}/ai/v1`
-}
-
-export const route = Route.make({
-  id: "cloudflare-workers-ai",
-  provider: id,
-  providerMetadataKey: "cloudflare-workers-ai",
-  protocol: OpenAIChat.protocol,
-  endpoint: Endpoint.path("/chat/completions"),
-  framing: OpenAIChat.framing,
-})
-
-export const routes = [route]
-
-export const configure = (input: LanguageModelOptions) => {
-  const { accountId: _accountId, apiKey: _apiKey, auth: _auth, baseURL: _baseURL, ...defaults } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: baseURL(input) },
-    auth: AuthOptions.bearer(input, authEnvVars),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) => configured.model<OpenAIProviderOptionsInput>({ id: modelID }),
-    configure,
-  }
-}
-
-export const provider = { id, configure }
-
-export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: baseURL(settings),
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
-
-export * as CloudflareWorkersAI from "./cloudflare-workers-ai.js"

packages/ai/src/providers/cloudflare.ts

@@ -0,0 +1,133 @@
+import type { Config, Redacted } from "effect"
+import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat.js"
+import { Auth } from "../route/auth.js"
+import { AuthOptions, type AtLeastOne, type ProviderAuthOption } from "../route/auth-options.js"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import type { OpenAIProviderOptionsInput } from "./openai-options.js"
+
+export const aiGatewayID = ProviderID.make("cloudflare-ai-gateway")
+export const workersAIID = ProviderID.make("cloudflare-workers-ai")
+export const aiGatewayAuthEnvVars = ["CLOUDFLARE_API_TOKEN", "CF_AIG_TOKEN"] as const
+export const workersAIAuthEnvVars = ["CLOUDFLARE_API_KEY", "CLOUDFLARE_WORKERS_AI_TOKEN"] as const
+
+type CloudflareSecret = string | Redacted.Redacted | Config.Config<string | Redacted.Redacted>
+
+type GatewayURL = AtLeastOne<{
+  readonly accountId: string
+  readonly baseURL: string
+}> & {
+  readonly gatewayId?: string
+}
+
+export type AIGatewayOptions = GatewayURL &
+  Omit<RouteDefaultsInput, "providerOptions"> &
+  ProviderAuthOption<"optional"> & {
+    /** Cloudflare AI Gateway authentication token. Sent as `cf-aig-authorization`. */
+    readonly gatewayApiKey?: CloudflareSecret
+    readonly providerOptions?: OpenAIProviderOptionsInput
+  }
+
+type WorkersAIURL = AtLeastOne<{
+  readonly accountId: string
+  readonly baseURL: string
+}>
+
+export type WorkersAIOptions = WorkersAIURL &
+  Omit<RouteDefaultsInput, "providerOptions"> &
+  ProviderAuthOption<"optional"> & {
+    readonly providerOptions?: OpenAIProviderOptionsInput
+  }
+
+export const aiGatewayBaseURL = (input: GatewayURL) => {
+  if (input.baseURL) return input.baseURL
+  if (!input.accountId) throw new Error("CloudflareAIGateway.configure requires accountId unless baseURL is supplied")
+  return `https://gateway.ai.cloudflare.com/v1/${encodeURIComponent(input.accountId)}/${encodeURIComponent(input.gatewayId?.trim() || "default")}/compat`
+}
+
+const aiGatewayAuth = (input: AIGatewayOptions) => {
+  if ("auth" in input && input.auth) return input.auth
+  const gateway = Auth.optional(input.gatewayApiKey, "gatewayApiKey")
+    .orElse(Auth.config("CLOUDFLARE_API_TOKEN"))
+    .orElse(Auth.config("CF_AIG_TOKEN"))
+    .pipe(Auth.bearerHeader("cf-aig-authorization"))
+  if (!("apiKey" in input) || input.apiKey === undefined) return gateway
+  if (input.gatewayApiKey === undefined) return Auth.bearer(input.apiKey)
+  return Auth.bearerHeader("cf-aig-authorization", input.gatewayApiKey).andThen(Auth.bearer(input.apiKey))
+}
+
+export const workersAIBaseURL = (input: WorkersAIURL) => {
+  if (input.baseURL) return input.baseURL
+  if (!input.accountId) throw new Error("CloudflareWorkersAI.configure requires accountId unless baseURL is supplied")
+  return `https://api.cloudflare.com/client/v4/accounts/${encodeURIComponent(input.accountId)}/ai/v1`
+}
+
+const workersAIAuth = (input: WorkersAIOptions) => {
+  return AuthOptions.bearer(input, workersAIAuthEnvVars)
+}
+
+export const aiGatewayRoute = OpenAICompatibleChat.route.with({
+  id: "cloudflare-ai-gateway",
+  provider: aiGatewayID,
+})
+
+export const workersAIRoute = OpenAICompatibleChat.route.with({
+  id: "cloudflare-workers-ai",
+  provider: workersAIID,
+})
+
+export const routes = [aiGatewayRoute, workersAIRoute]
+
+const aiGatewayDefaults = (options: AIGatewayOptions) => {
+  const {
+    accountId: _accountId,
+    gatewayId: _gatewayId,
+    apiKey: _apiKey,
+    gatewayApiKey: _gatewayApiKey,
+    baseURL: _baseURL,
+    auth: _auth,
+    ...rest
+  } = options
+  return rest
+}
+
+const workersAIDefaults = (options: WorkersAIOptions) => {
+  const { accountId: _accountId, apiKey: _apiKey, auth: _auth, baseURL: _baseURL, ...rest } = options
+  return rest
+}
+
+const configureAIGateway = (options: AIGatewayOptions) => {
+  const route = aiGatewayRoute.with({
+    ...aiGatewayDefaults(options),
+    endpoint: { baseURL: aiGatewayBaseURL(options) },
+    auth: aiGatewayAuth(options),
+  })
+  return {
+    id: aiGatewayID,
+    model: (modelID: string | ModelID) => route.model<OpenAIProviderOptionsInput>({ id: modelID }),
+    configure: configureAIGateway,
+  }
+}
+
+const configureWorkersAI = (options: WorkersAIOptions) => {
+  const route = workersAIRoute.with({
+    ...workersAIDefaults(options),
+    endpoint: { baseURL: workersAIBaseURL(options) },
+    auth: workersAIAuth(options),
+  })
+  return {
+    id: workersAIID,
+    model: (modelID: string | ModelID) => route.model<OpenAIProviderOptionsInput>({ id: modelID }),
+    configure: configureWorkersAI,
+  }
+}
+
+export const CloudflareAIGateway = {
+  id: aiGatewayID,
+  configure: configureAIGateway,
+}
+
+export const CloudflareWorkersAI = {
+  id: workersAIID,
+  configure: configureWorkersAI,
+}

packages/ai/src/providers/deepinfra.ts

@@ -1,13 +1,12 @@
 import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
+import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
+import type { RouteDefaultsInput } from "../route/client.js"
 import { ProviderID, type ModelID } from "../schema/index.js"
+import { profiles } from "./openai-compatible-profile.js"
 import type { OpenAIProviderOptionsInput } from "./openai-options.js"
 
 export const id = ProviderID.make("deepinfra")
-const baseURL = "https://api.deepinfra.com/v1/openai"
 
 export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
   ProviderAuthOption<"optional"> & {
@@ -21,24 +20,21 @@ export interface Settings extends ProviderPackage.Settings {
   readonly providerOptions?: OpenAIProviderOptionsInput
 }
 
-export const route = Route.make({
+export const route = OpenAICompatibleChat.route.with({
   id: "deepinfra-chat",
   provider: id,
-  providerMetadataKey: "deepinfra",
-  protocol: OpenAIChat.protocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL }),
-  framing: OpenAIChat.framing,
+  endpoint: { baseURL: profiles.deepinfra.baseURL },
 })
 
 export const routes = [route]
 
 export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL: endpoint, ...defaults } = input
-  const root = endpoint?.replace(/\/+$/, "")
+  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
+  const root = baseURL?.replace(/\/+$/, "")
   const configured = route.with({
     ...defaults,
     endpoint: {
-      baseURL: root === undefined ? baseURL : root.endsWith("/openai") ? root : `${root}/openai`,
+      baseURL: root === undefined ? profiles.deepinfra.baseURL : root.endsWith("/openai") ? root : `${root}/openai`,
     },
     auth: AuthOptions.bearer(input, "DEEPINFRA_API_KEY"),
   })

packages/ai/src/providers/deepseek.ts

@@ -1,64 +0,0 @@
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { ProviderID, type ModelID } from "../schema/index.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("deepseek")
-const baseURL = "https://api.deepseek.com/v1"
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: OpenAIProviderOptionsInput
-}
-
-export const route = Route.make({
-  id: "deepseek-chat",
-  provider: id,
-  providerMetadataKey: "deepseek",
-  protocol: OpenAIChat.protocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL }),
-  framing: OpenAIChat.framing,
-})
-
-export const routes = [route]
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL: endpoint, ...defaults } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: endpoint ?? baseURL },
-    auth: AuthOptions.bearer(input, "DEEPSEEK_API_KEY"),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) =>
-      configured.model<OpenAIProviderOptionsInput>({
-        id: modelID,
-        compatibility: { maxTokensField: "max_tokens", supportsStore: false },
-      }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
-
-export * as DeepSeek from "./deepseek.js"

packages/ai/src/providers/fireworks.ts

@@ -1,60 +0,0 @@
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { ProviderID, type ModelID } from "../schema/index.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("fireworks")
-const baseURL = "https://api.fireworks.ai/inference/v1"
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: OpenAIProviderOptionsInput
-}
-
-export const route = Route.make({
-  id: "fireworks-chat",
-  provider: id,
-  providerMetadataKey: "fireworks",
-  protocol: OpenAIChat.protocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL }),
-  framing: OpenAIChat.framing,
-})
-
-export const routes = [route]
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL: endpoint, ...defaults } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: endpoint ?? baseURL },
-    auth: AuthOptions.bearer(input, "FIREWORKS_API_KEY"),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) => configured.model<OpenAIProviderOptionsInput>({ id: modelID }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
-
-export * as Fireworks from "./fireworks.js"

packages/ai/src/providers/google-vertex-chat.ts

@@ -1,7 +1,6 @@
 import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
+import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
+import type { RouteDefaultsInput } from "../route/client.js"
 import { ProviderID, type ModelID } from "../schema/index.js"
 import { GoogleVertexShared } from "./google-vertex-shared.js"
 import type { OpenAIProviderOptionsInput } from "./openai-options.js"
@@ -25,13 +24,10 @@ export interface Settings extends ProviderPackage.Settings {
   readonly providerOptions?: OpenAIProviderOptionsInput
 }
 
-const route = Route.make({
+const route = OpenAICompatibleChat.route.with({
   id: "google-vertex-chat",
   provider: id,
   providerMetadataKey: "vertex",
-  protocol: OpenAIChat.protocol,
-  endpoint: Endpoint.path("/chat/completions"),
-  framing: OpenAIChat.framing,
 })
 
 export const routes = [route]

packages/ai/src/providers/google-vertex-messages.ts

@@ -57,9 +57,7 @@ const route = Route.make({
   }),
   endpoint: Endpoint.path(({ request }) => `/${request.model.id}:streamRawPredict`),
   auth: Auth.none,
-  transport: AnthropicMessages.transport<
-    Omit<AnthropicMessages.AnthropicMessagesBody, "model"> & { readonly anthropic_version: typeof VERSION }
-  >(),
+  framing: AnthropicMessages.framing,
   headers: () => ({ "anthropic-version": HEADER_VERSION }),
 })
 

packages/ai/src/providers/google-vertex-responses.ts

@@ -1,7 +1,6 @@
 import type { ProviderPackage } from "../provider-package.js"
-import { OpenResponses } from "../protocols/open-responses.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
+import { OpenAICompatibleResponses } from "../protocols/openai-compatible-responses.js"
+import type { RouteDefaultsInput } from "../route/client.js"
 import { ProviderID, type ModelID } from "../schema/index.js"
 import { GoogleVertexShared } from "./google-vertex-shared.js"
 import type { OpenResponsesProviderOptionsInput } from "./open-responses-options.js"
@@ -25,14 +24,11 @@ export interface Settings extends ProviderPackage.Settings {
   readonly providerOptions?: OpenResponsesProviderOptionsInput
 }
 
-const route = Route.make({
+const route = OpenAICompatibleResponses.route.with({
   id: "google-vertex-responses",
   provider: id,
   providerMetadataKey: "vertex",
-  protocol: OpenResponses.protocol,
-  endpoint: Endpoint.path(OpenResponses.PATH),
-  transport: OpenResponses.httpTransport,
-  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
+  providerOptions: { store: false },
 })
 
 export const routes = [route]

packages/ai/src/providers/groq.ts

@@ -5,12 +5,13 @@ import { ProviderShared } from "../protocols/shared.js"
 import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
 import { Route, type RouteDefaultsInput } from "../route/client.js"
 import { Endpoint } from "../route/endpoint.js"
+import { Framing } from "../route/framing.js"
 import { Protocol } from "../route/protocol.js"
 import { ProviderID, type ModelID, type LLMRequest } from "../schema/index.js"
+import { profiles } from "./openai-compatible-profile.js"
 import type { OpenAIProviderOptionsInput } from "./openai-options.js"
 
 export const id = ProviderID.make("groq")
-const baseURL = "https://api.groq.com/openai/v1"
 
 export type ProviderOptions = Pick<OpenAIProviderOptionsInput, "reasoningEffort"> & {
   /** Controls visible reasoning on GPT-OSS; other models always use parsed reasoning. */
@@ -73,15 +74,15 @@ export const route = Route.make({
   provider: id,
   providerMetadataKey: "openai",
   protocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL }),
-  framing: OpenAIChat.framing,
+  endpoint: Endpoint.path("/chat/completions", { baseURL: profiles.groq.baseURL }),
+  framing: Framing.sse,
 })
 
 export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL: endpoint, ...defaults } = input
+  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
   const configured = route.with({
     ...defaults,
-    endpoint: { baseURL: endpoint ?? baseURL },
+    endpoint: { baseURL: baseURL ?? profiles.groq.baseURL },
     auth: AuthOptions.bearer(input, "GROQ_API_KEY"),
   })
   return {

packages/ai/src/providers/index.ts

@@ -3,22 +3,16 @@ export * as AnthropicCompatible from "./anthropic-compatible.js"
 export * as AmazonBedrock from "./amazon-bedrock.js"
 export * as AmazonBedrockMantle from "./amazon-bedrock-mantle.js"
 export * as Azure from "./azure.js"
-export * as Baseten from "./baseten.js"
 export * as Cerebras from "./cerebras.js"
-export * as CloudflareAIGateway from "./cloudflare-ai-gateway.js"
-export * as CloudflareWorkersAI from "./cloudflare-workers-ai.js"
+export * as Cloudflare from "./cloudflare.js"
+export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare.js"
 export * as DeepInfra from "./deepinfra.js"
-export * as DeepSeek from "./deepseek.js"
-export * as Fireworks from "./fireworks.js"
 export * as Google from "./google.js"
 export * as GoogleVertex from "./google-vertex.js"
 export * as GoogleVertexChat from "./google-vertex-chat.js"
 export * as GoogleVertexMessages from "./google-vertex-messages.js"
 export * as GoogleVertexResponses from "./google-vertex-responses.js"
 export * as Groq from "./groq.js"
-export * as Meta from "./meta.js"
-export * as MiniMax from "./minimax.js"
-export * as Mistral from "./mistral.js"
 export * as OpenAI from "./openai.js"
 export * as OpenAICompatible from "./openai-compatible.js"
 export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"

packages/ai/src/providers/meta.ts

@@ -1,182 +0,0 @@
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { AnthropicMessages } from "../protocols/anthropic-messages.js"
-import { MetaResponses } from "../protocols/meta-responses.js"
-import { MetaMessages } from "../protocols/meta-messages.js"
-import { MetaImages } from "../protocols/meta-images.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { HttpOptions, ProviderID, ToolDefinition, type ModelID } from "../schema/index.js"
-import type { OpenResponsesProviderOptionsInput } from "./open-responses-options.js"
-
-export const id = ProviderID.make("meta")
-const baseURL = "https://api.meta.ai/v1"
-
-export type ProviderOptionsInput = OpenResponsesProviderOptionsInput &
-  Pick<AnthropicMessages.OptionsInput, "thinking" | "effort">
-export type MessagesOptionsInput = Pick<
-  AnthropicMessages.OptionsInput,
-  "thinking" | "effort" | "outputConfig" | "output_config" | "serviceTier" | "service_tier" | "metadata"
-> & { readonly [key: string]: unknown }
-export type ImageOptions = MetaImages.ImageOptions
-
-export interface WebSearchOptions {
-  readonly searchContextSize?: "low" | "medium" | "high" | (string & {})
-  readonly userLocation?: {
-    readonly city?: string
-    readonly region?: string
-    readonly country?: string
-    readonly timezone?: string
-  }
-}
-
-export const webSearch = (options: WebSearchOptions = {}) =>
-  ToolDefinition.make({
-    name: "web_search",
-    description: "Search the web with Meta's hosted search tool.",
-    inputSchema: { type: "object", properties: {}, additionalProperties: false },
-    native: {
-      meta: {
-        type: "web_search",
-        search_context_size: options.searchContextSize,
-        user_location:
-          options.userLocation === undefined ? undefined : { type: "approximate", ...options.userLocation },
-      },
-    },
-  })
-
-export interface ImageGenerationOptions {
-  readonly size?: string
-  readonly outputFormat?: "webp" | "png" | "jpeg" | (string & {})
-  readonly reasoningStrength?: "low" | "high" | (string & {})
-  readonly enableImageSearch?: boolean
-  readonly enableWebSearch?: boolean
-  readonly enableShell?: boolean
-}
-
-export const imageGeneration = (options: ImageGenerationOptions = {}) =>
-  ToolDefinition.make({
-    name: "image_generation",
-    description: "Generate or edit an image with Muse Image.",
-    inputSchema: { type: "object", properties: {}, additionalProperties: false },
-    native: {
-      meta: {
-        type: "image_generation",
-        size: options.size,
-        output_format: options.outputFormat,
-        reasoning_strength: options.reasoningStrength,
-        enable_image_search: options.enableImageSearch,
-        enable_web_search: options.enableWebSearch,
-        enable_shell: options.enableShell,
-      },
-    },
-  })
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: ProviderOptionsInput
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: ProviderOptionsInput
-}
-
-const responsesRoute = Route.make({
-  id: "meta-responses",
-  provider: id,
-  providerMetadataKey: "meta",
-  protocol: MetaResponses.protocol,
-  endpoint: Endpoint.path("/responses", { baseURL }),
-  // Meta Responses does not support WebSocket upgrades; always use HTTP/SSE.
-  transport: MetaResponses.httpTransport,
-  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
-})
-
-const chatRoute = Route.make({
-  id: "meta-chat",
-  provider: id,
-  providerMetadataKey: "meta",
-  protocol: OpenAIChat.protocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL }),
-  framing: OpenAIChat.framing,
-})
-
-const messagesRoute = Route.make({
-  id: "meta-messages",
-  provider: id,
-  providerMetadataKey: "meta",
-  protocol: MetaMessages.protocol,
-  endpoint: Endpoint.path("/messages", { baseURL }),
-  framing: AnthropicMessages.framing,
-  defaults: { providerOptions: { thinking: { type: "adaptive", display: "omitted" } } },
-})
-
-export const routes = [responsesRoute, chatRoute, messagesRoute]
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL: endpoint, ...defaults } = input
-  const options = {
-    ...defaults,
-    endpoint: { baseURL: endpoint ?? baseURL },
-    auth: AuthOptions.bearer(input, "META_API_KEY"),
-  }
-  const configuredResponses = responsesRoute.with(options)
-  const configuredChat = chatRoute.with(options)
-  const configuredMessages = messagesRoute.with(options)
-  const responses = (modelID: string | ModelID) =>
-    configuredResponses.model<OpenResponsesProviderOptionsInput>({ id: modelID })
-  const chat = (modelID: string | ModelID) =>
-    configuredChat.model<OpenResponsesProviderOptionsInput>({
-      id: modelID,
-      compatibility: { maxTokensField: "max_completion_tokens", supportsStore: false },
-    })
-  const messages = (modelID: string | ModelID) =>
-    configuredMessages.model<MessagesOptionsInput>({
-      id: modelID,
-      compatibility: { requireSignature: false },
-    })
-  const image = (modelID: string | ModelID) =>
-    MetaImages.model({
-      id: modelID,
-      baseURL: endpoint ?? baseURL,
-      auth: options.auth,
-      headers: input.headers,
-      http: input.http === undefined ? undefined : HttpOptions.make(input.http),
-    })
-  return { id, model: responses, responses, chat, messages, image, configure }
-}
-
-export const provider = configure()
-export const responses = provider.responses
-export const chat = provider.chat
-export const messages = provider.messages
-export const image = provider.image
-
-export const model: ProviderPackage.Definition<Settings, OpenResponsesProviderOptionsInput>["model"] = (
-  modelID,
-  settings,
-) => fromSettings(settings).responses(modelID)
-
-export const chatModel: ProviderPackage.Definition<Settings, OpenResponsesProviderOptionsInput>["model"] = (
-  modelID,
-  settings,
-) => fromSettings(settings).chat(modelID)
-
-export const messagesModel: ProviderPackage.Definition<Settings, MessagesOptionsInput>["model"] = (modelID, settings) =>
-  fromSettings(settings).messages(modelID)
-
-function fromSettings(settings: Settings) {
-  return configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers,
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  })
-}
-
-export * as Meta from "./meta.js"

packages/ai/src/providers/meta/chat.ts

@@ -1,2 +0,0 @@
-export { chatModel as model } from "../meta.js"
-export type { Settings } from "../meta.js"

packages/ai/src/providers/meta/messages.ts

@@ -1,2 +0,0 @@
-export { messagesModel as model } from "../meta.js"
-export type { Settings } from "../meta.js"

packages/ai/src/providers/meta/responses.ts

@@ -1,2 +0,0 @@
-export { model } from "../meta.js"
-export type { Settings } from "../meta.js"

packages/ai/src/providers/minimax.ts

@@ -1,144 +0,0 @@
-import { Effect, Schema } from "effect"
-import type { ProviderPackage } from "../provider-package.js"
-import { AnthropicMessages } from "../protocols/anthropic-messages.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { OpenResponses } from "../protocols/open-responses.js"
-import { ProviderShared } from "../protocols/shared.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { Framing } from "../route/framing.js"
-import { Protocol } from "../route/protocol.js"
-import { ProviderID, type LLMRequest, type ModelID } from "../schema/index.js"
-
-export const id = ProviderID.make("minimax")
-
-export type MessagesOptionsInput = {
-  /** M3 defaults to disabled; M2.x always thinks. */
-  readonly thinking?: { readonly type: "adaptive" | "disabled" }
-  readonly metadata?: AnthropicMessages.OptionsInput["metadata"]
-}
-
-export type ChatOptionsInput = {
-  /** M3 defaults to adaptive; M2.x always thinks. */
-  readonly thinking?: { readonly type: "adaptive" | "disabled" | (string & {}) }
-  /** Separates reasoning from text. Defaults to true. */
-  readonly reasoningSplit?: boolean
-}
-
-export type ResponsesOptionsInput = {
-  /** M3 defaults to none. Other supported values enable thinking without changing its depth. */
-  readonly reasoningEffort?: "none" | "minimal" | "low" | "medium" | "high" | (string & {})
-}
-
-export type ProviderOptionsInput = MessagesOptionsInput | ChatOptionsInput | ResponsesOptionsInput
-
-export type Config = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    /** Overrides the selected API's base URL, including its version prefix. */
-    readonly baseURL?: string
-    readonly providerOptions?: ProviderOptionsInput
-  }
-
-export interface Settings<Options = MessagesOptionsInput> extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: Options
-}
-
-const ChatOptions = Schema.Struct({
-  thinking: Schema.optional(Schema.Struct({ type: Schema.String })),
-  reasoningSplit: Schema.optional(Schema.Boolean),
-})
-
-const chatProtocol = Protocol.make({
-  id: "minimax-chat",
-  body: {
-    schema: Schema.Struct({
-      ...OpenAIChat.bodyFields,
-      thinking: ChatOptions.fields.thinking,
-      reasoning_split: Schema.Boolean,
-    }),
-    from: Effect.fn("MiniMax.chatFromRequest")(function* (request: LLMRequest) {
-      const options = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(ChatOptions))(
-        request.providerOptions ?? {},
-      )
-      return {
-        ...(yield* OpenAIChat.protocol.body.from(request)),
-        thinking: options.thinking,
-        // MiniMax otherwise embeds <think> tags in ordinary assistant text.
-        reasoning_split: options.reasoningSplit ?? true,
-      }
-    }),
-  },
-  stream: OpenAIChat.protocol.stream,
-})
-
-const messagesRoute = Route.make({
-  id: "minimax-messages",
-  provider: id,
-  providerMetadataKey: "minimax",
-  protocol: AnthropicMessages.protocol,
-  endpoint: Endpoint.path("/messages", { baseURL: "https://api.minimax.io/anthropic/v1" }),
-  framing: AnthropicMessages.framing,
-  headers: () => ({ "anthropic-version": "2023-06-01" }),
-})
-
-const chatRoute = Route.make({
-  id: "minimax-chat",
-  provider: id,
-  providerMetadataKey: "minimax",
-  protocol: chatProtocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL: "https://api.minimax.io/v1" }),
-  framing: OpenAIChat.framing,
-})
-
-const responsesRoute = Route.make({
-  id: "minimax-responses",
-  provider: id,
-  providerMetadataKey: "minimax",
-  protocol: OpenResponses.protocol,
-  endpoint: Endpoint.path("/responses", { baseURL: "https://api.minimax.io/v1" }),
-  framing: Framing.sse,
-})
-
-export const routes = [messagesRoute, chatRoute, responsesRoute]
-
-export const configure = (input: Config = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL, ...rest } = input
-  const defaults = {
-    ...rest,
-    endpoint: baseURL === undefined ? undefined : { baseURL },
-    auth: AuthOptions.bearer(input, "MINIMAX_API_KEY"),
-  }
-  const messages = (modelID: string | ModelID) =>
-    messagesRoute.with(defaults).model<MessagesOptionsInput>({ id: modelID })
-  const chat = (modelID: string | ModelID) =>
-    chatRoute.with(defaults).model<ChatOptionsInput>({
-      id: modelID,
-      compatibility: { supportsStore: false, supportsStrictMode: false },
-    })
-  const responses = (modelID: string | ModelID) =>
-    responsesRoute.with(defaults).model<ResponsesOptionsInput>({ id: modelID })
-  return { id, model: messages, messages, chat, responses, configure }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings<MessagesOptionsInput>, MessagesOptionsInput>["model"] = (
-  modelID,
-  settings,
-) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers,
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
-
-export const messages = provider.messages
-export const chat = provider.chat
-export const responses = provider.responses
-
-export * as MiniMax from "./minimax.js"

packages/ai/src/providers/minimax/chat.ts

@@ -1,13 +0,0 @@
-import type { ProviderPackage } from "../../provider-package.js"
-import { MiniMax } from "../minimax.js"
-
-export type Settings = MiniMax.Settings<MiniMax.ChatOptionsInput>
-
-export const model: ProviderPackage.Definition<Settings, MiniMax.ChatOptionsInput>["model"] = (modelID, settings) =>
-  MiniMax.configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers,
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).chat(modelID)

packages/ai/src/providers/minimax/messages.ts

@@ -1 +0,0 @@
-export { model, type Settings, type MessagesOptionsInput } from "../minimax.js"

packages/ai/src/providers/minimax/responses.ts

@@ -1,16 +0,0 @@
-import type { ProviderPackage } from "../../provider-package.js"
-import { MiniMax } from "../minimax.js"
-
-export type Settings = MiniMax.Settings<MiniMax.ResponsesOptionsInput>
-
-export const model: ProviderPackage.Definition<Settings, MiniMax.ResponsesOptionsInput>["model"] = (
-  modelID,
-  settings,
-) =>
-  MiniMax.configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers,
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).responses(modelID)

packages/ai/src/providers/mistral.ts

@@ -1,51 +0,0 @@
-import type { ProviderPackage } from "../provider-package.js"
-import { MistralChat } from "../protocols/mistral-chat.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import type { RouteDefaultsInput } from "../route/client.js"
-import { ProviderID, type ModelID } from "../schema/index.js"
-
-export const id = ProviderID.make("mistral")
-
-export type ProviderOptions = MistralChat.ProviderOptionsInput
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: ProviderOptions
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: ProviderOptions
-}
-
-export const route = MistralChat.route
-export const routes = [route]
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: baseURL ?? MistralChat.DEFAULT_BASE_URL },
-    auth: AuthOptions.bearer(input, "MISTRAL_API_KEY"),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) => configured.model<ProviderOptions>({ id: modelID }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, ProviderOptions>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
-
-export * as Mistral from "./mistral.js"